Datasets:

---

ExponentialScience

/

DLT-Scientific-Literature

like 0

Follow

Exponential Science 4

ERROR: type should be string, got "https://doi.org/10.1007/s10489 020 01664 w\n\n# A general-purpose distributed pattern mining system\n\nAsma Belhadi [1] & Youcef Djenouri [2,3] & Jerry Chun-Wei Lin [4] & Alberto Cano [5]\n\nPublished online: 18 March 2020\n\n# The Author(s) 2020\n\nAbstract\nThis paper explores five pattern mining problems and proposes a new distributed framework called DT-DPM: Decomposition\nTransaction for Distributed Pattern Mining. DT-DPM addresses the limitations of the existing pattern mining problems by\nreducing the enumeration search space. Thus, it derives the relevant patterns by studying the different correlation among the\ntransactions. It first decomposes the set of transactions into several clusters of different sizes, and then explores heterogeneous\narchitectures, including MapReduce, single CPU, and multi CPU, based on the densities of each subset of transactions. To\nevaluate the DT-DPM framework, extensive experiments were carried out by solving five pattern mining problems (FIM:\nFrequent Itemset Mining, WIM: Weighted Itemset Mining, UIM: Uncertain Itemset Mining, HUIM: High Utility Itemset\nMining, and SPM: Sequential Pattern Mining). Experimental results reveal that by using DT-DPM, the scalability of the pattern\nmining algorithms was improved on large databases. Results also reveal that DT-DPM outperforms the baseline parallel pattern\nmining algorithms on big databases.\n\nKeywords Pattern mining [. ]Decomposition [. ]Distributed computing [. ]Heterogeneous architecture\n\n\n### 1 Introduction\n\nPattern mining is a data mining task that aims at studying the\ncorrelations within data and discovering relevant patterns\nfrom large databases. In practice, different database representation could be observed (from Boolean databases to sequence\ndatabases). The problem of pattern mining is to find an efficient approach to extract the relevant patterns in a database. It\n\n- Asma Belhadi\n[abelhadi@usthb.dz](mailto:abelhadi@usthb.dz)\n\nYoucef Djenouri\nyoucef.djenouri@sintef.no\n\nJerry Chun-Wei Lin\njerrylin@ieee.org\n\nAlberto Cano\nacano@vcu.edu\n\n1 Department of Computer Science, USTHB, Algiers, Algeria\n\n2 Department of Computer Science, NTNU, Trondheim, Norway\n\n3 SINTEF Digital, Forskningsveien 1, 0314 Oslo, Norway\n\n4 Department of Computing, Mathematics and Physics, Western\nNorway University of Applied Sciences (HVL), Bergen, Norway\n\n5 Department of Computer Science, Virginia Commonwealth\nUniversity, Richmond, VA, USA\n\n\nis used in many applications and domains such as ontology\nmatching [1], process mining [2], decision making [3], and\nconstraint programming [4]. The pattern mining is also called\nwith “Big data” applications such as in frequent genes extractions from DNA in Bio-informatics [5], relevant hashtags\nfrom twitter streams in social network analysis [6], analysis\nof sensorial data from IoT devices in smart city applications\n\n[7]. This work mainly focuses on mining the information from\nbig transactional databases.\n\n#### 1.1 Motivation\n\nSolutions to pattern mining problems [8–12] are high time\nconsuming when dealing with large and very large databases\nfor pattern mining problems such as FIM and WIM, and they\nare totally inefficient when solving more complex problems\nsuch as UIM, HUIM, and SPM. To improve the runtime performance of the pattern mining approaches, many optimization and high performance computing techniques have been\nproposed [13–18]. However, these strategies are inefficient\nwhen dealing with big databases, where only few number of\nrelevant patterns are useful and displayed to the end user. We\ncontemplate that these algorithms are inefficient because they\nconsider the whole database in the mining process. In our\nprevious work [19], we proposed a new algorithm for pattern\nmining algorithm, where the aim is to study the correlation\n\n\n-----\n\nbetween the input data to split the whole problem into many\nsmaller sub-problems, but as being as independent as possible.\nWe proposed a k-means algorithm to assign the transactions\ninto different clusters. We also developed an efficient strategy\nto accurately explore the clusters of transactions. This approach gives good results compared to the baseline serial\nmethods. However, it still suffers from the runtime and accuracy performance when dealing with big databases. This is\ndue to the separator items between clusters, where the mining\nprocess of these items should be carried out by exploring the\ntransactions of all clusters. This issue degrades the overall\nperformance of such an approach. Motivated by the preliminary results reported in [19], we propose a new parallel framework, which addresses the following issues, i) minimizing the\nnumber of separator items, ii) improving the runtime and accuracy on big databases.\n\n#### 1.2 Contributions\n\nIn this research work, we propose a generic intelligent pattern\nmining algorithm for dealing pattern mining problems on big\ndatabases. It is a comprehensive extension of our previous\nwork [19]. With this in mind, the main contributions of this\nwork are as follows:\n\n1. Propose a new framework called DT-DPM for improving\npattern mining algorithms in a distributed environment.\n2. Develop a decomposition approach to cluster the transactions set into smaller similar groups.\n3. Extend the MapReduce computing framework to deal\nwith the pattern mining algorithms by exploiting the different dependencies between the transactions of the\nclusters.\n4. Five cases studies (FIM, WIM, UIM, HUIM, and SPM)\nhave been analyzed on well-known pattern mining databases by considering five best pattern mining algorithms\nin terms of time complexity as baseline algorithms for the\nDT-DPM framework. Experimental results reveal that by\nusing DT-DPM, the scalability of the pattern mining algorithms was improved on large databases. Results also\nreveal that DT-DPM outperforms the baseline parallel pattern mining algorithms on big databases.\n\n#### 1.3 Outline\n\nThe remainder of the paper is organized as follows: Section 2\nintroduces the basic concepts of pattern mining problems.\nSection 3 reviews existing pattern mining algorithms followed\nby a detailed explanation of our DT-DPM framework in\nSection 4. The performance evaluation is presented in\nSection 5 whereas Section 6 draws the conclusions.\n\n\n### 2 Pattern mining problems\n\nIn this section, we first present a general formulation of pattern\nmining and then we present a few pattern mining problems\naccording to the general formulation.\n\nDefinition 1 (pattern) Let us consider I = {1, 2, …, n} as a set\nof items where n is the number of items, and T = {t1, t2, dots,\ntm} as a set of transactions where m is the number of transactions. We define the function a, where for the item i in the\ntransaction tj, the corresponding pattern reads p=σ(i, j).\n\nDefinition 2 (pattern mining) A pattern mining problem finds\nthe set of all relevant patterns L, such as\n\nL p Interestingness T ; I; p ≥ γ 1\n¼ f j ð Þ g ð Þ\n\nwhere the Interestigness(T, I, p) is the measure to evaluate a\npattern p among the set of transactions T and the set of items I,\nwhere γ is the mining threshold.\nFrom these two definitions, we present the existing pattern\nmining problems.\n\nDefinition 3 (Boolean database) We define a Boolean database\nby setting the function σ (see Def. 2) as\n\n� 1 if i∈t j\n\nσ i; j 2\nð Þ ¼ 0 otherwise ð Þ\n\nDefinition 4 (frequent itemset mining (FIM)) We define a FIM\nproblem as an extension of the pattern mining problem (see\nDef. 2) by\n\nL p Support T ; I; p ≥ γ 3\n¼ f j ð Þ g ð Þ\n\nwith Support Tð ; I; pÞ ¼ j jpjTTj;I [where][ T][ is the set of transactions]\n\nin a Boolean database defined by Def. 1, γ is a minimum\nsupport threshold, and |p|T, I is the number of transactions in\nT containing the pattern p.\n\n\nDefinition 5 (weighted database) We define a weighted database by setting the function σ (see Def. 2) as\n\n� wij if i∈t j\n\nσ i; j 4\nð Þ ¼ 0 otherwise ð Þ\n\nNote that wij is the weight of the item i in the transaction tj.\n\nDefinition 6 (weighted itemset mining (WIM)) We define a\nWIM problem as an extension of the pattern mining problem\n(see Def. 2) by\n\nL p WS T ; I; p ≥ γ 5\n¼ f j ð Þ g ð Þ\n\n\n-----\n\nwith WS Tð ; I; pÞ ¼ ∑[j]j[T]¼[j]1[W t]� [ j][;][ I][;][ p]� where T is the set of\ntransactions in the weighted database defined by Def. 3,\nW(tj, I, p) is the minimum weight of the items of the pattern\np in the transaction tj, and γ is a minimum weighted support\nthreshold.\n\nDefinition 7 (uncertain database) We define an uncertain database by setting the function σ (see Def. 2) as\n\n� Probij if i∈t j\n\nσ i; j 6\nð Þ ¼ 0 otherwise ð Þ\n\nNote that Probij is the uncertainty value of i in the transaction tj.\n\nDefinition 8 (uncertain itemset mining (UIM)) We define a\nUIM problem as an extension of the pattern mining problem\n(see Def. 2) by\n\nL p US T ; I; p ≥ γ 7\n¼ f j ð Þ g ð Þ\n\nwith US Tð ; I; pÞ ¼ ∑[j]j[T]¼[j]1[∏]i∈p[Prob][ij][ where][ T][ is the set of]\ntransactions in the uncertain database defined by Def. 5 and\nγ is the minimum uncertain support threshold.\n\nDefinition 9 (utility database) We define an utility database by\nsetting the function σ (see Def. 2) as\n\n� iuij if i∈t j\n\nσ i; j 8\nð Þ ¼ 0 otherwise ð Þ\n\nNote that iuij is the internal utility value of i in the transaction tj, we also define external utility of each item i by eu(i).\n\nDefinition 10 (high utility itemset mining (HUIM)) We define a\nHUIM problem as an extension of the pattern mining problem\n(see Def. 2) by\n\nL p U T ; I; p ≥ γ ; with U T ; I; p\n¼ f j ð Þ g ð Þ\n\njT j\n¼ ∑j¼1 i∑∈p iuij � eu ið Þ ð9Þ\n\nwhere T is the set of transactions in the utility database defined\nby Def. 7 and γ is the minimum utility threshold.\n\nDefinition 11 (sequence database) We assume a total order on\nitems ≺, such as 1 ≺ 2 ≺ 3… ≺ n. A sequence is an ordered list\nof itemsets s = {I1, I2, …, I∣s∣}. Each itemset Ii is defined by\nsetting the function σ (see Def. 2) as σ(i, j) = i, if i ∈ tj\n\nDefinition 12 (sequential pattern mining (SPM)) We define a\nSPM problem as an extension of the pattern mining problem\n(see Def. 2) by\n\n\nL p Support T ; I; p ≥ γ 10\n¼ f j ð Þ g ð Þ\n\nwhere T is the set of transactions in the sequence database\ndefined by Def. 9 and γ is the minimum support threshold.\n\n### 3 Related work\n\nPattern mining has been largely studied in the last three decades [8–11, 20, 21]. There are many variants of pattern mining problem such as FIM, WIM, HUIM, UIM and SPM.\n\nFIM It aims at extracting all frequent itemsets that exceed the\nminimum support threshold. Apriori [22] and FP-Growth [23]\nare the most popular algorithms. Apriori applies a generate\nand test strategy to explore the itemset space. The candidate\nitemsets are generated incrementally and recursively. To generate k-sized itemsets as candidates, the algorithm calculates\nand combines the frequent (k-1)-sized itemsets. This process is\nrepeated until no candidate itemsets are obtained in an iteration. However, FP-Growth adopts a divide and conquer strategy and compresses the transactional database in the volatile\nmemory using an efficient tree structure. It then applies recursively the mining process to find the frequent itemsets. The\nmain limitation of the traditional FIM algorithms is the database format where only binary items can be mined. A typical\napplication of this problem is the market basket analysis,\nwhich a given item (product) may be present or absent in the\ngiven transaction (customer).\n\nWIM To address the FIM limitation, WIM is introduced, where\nweights are associated to each item to indicate their relative\nimportance in the given transaction [24]. The goal is to extract\nitemsets exceeding minimum weight threshold. The first WIM\nalgorithm is called WFIM: Weighted Frequent Itemset Mining\n\n[25]. It defines a weight range and a minimum weight constraint into the FP-Growth algorithm. Both weight and support\nmeasures are considered to prunethe search space. Yun [26]\nproposed WIP: Weighted Interesting Pattern. It introduces an\ninfinity measure that determines the correlation between the\nitems of the same pattern. The integration of the WIM in both\nApriori and FP-Growth is studied in [27]. The results showed\nthat FP-Growth outperforms Apriori for mining weighted patterns. Le et al. [28] proposed a frequent subgraph algorithm on\na weighted large graph. A novel strategy is developed which\naims to compute the weight of all candidate subgraphs. An\nefficient pruning strategy aims at reducing both the processing\ntime and the memory usage. Lee et al. [29] mine the frequent\nweighted itemsets by employing a novel type of prefix tree\nstructures. This allows to retrieve the relevant patterns more\naccurately without saving the list of identification number of\nthe different transactions.\n\n\n-----\n\nUIM An extension of WIM, called UIM, explores uncertain\ntransactional databases, where two models (expected-support\nand probabilistic itemsets) are defined to mine uncertain patterns. Li et al. [30] proposed the PFIMoS: Probabilistic\nFrequent Itemset Mining over Streams algorithm. It derives\nthe probabilistic frequent itemsets in an incremental way by\ndetermining the upper and the lower bounds of the mining\nthreshold. Lee et al. [31] introduced the U-WFI:\nUncertainmining of Weighted Frequent Itemsets algorithm.\nIt allows to discover from a given uncertain database relevant\nuncertain frequent itemsets with high probability values by\nfocusing on item weights. Liaqat et al. [32] show the use of\nuncertain frequent patterns in the image retrieval process. It\nincorporates the fuzzy ontology and uncertain frequent pattern\nmining for finding the relevant images regarding the user query. Lee et al. [33] suggest novel data structures to guarantee the\ncorrectness of the mining outputs without any false positives.\nIt allows to retrieve a complete set of uncertain relevant patterns in an reasonable amount of time.\n\nHUIM High Utility Itemset Mining is an extension of WIM\nwhere both internal and external utilities of the items are involved. The aim is to find all high utility patterns from transactional database that exceed the minimum utility threshold.\nThe utility of a pattern is the sum of the utility of all its items,\nwhere the utility of an item is defined by the product by its\ninternal and external utility values. Chan et al. [34] proposed\nthe first HUIM algorithm. It applies the Apriori-based algorithm to discover top k high utility patterns. This algorithm\nsuffers from the runtime performance, because the search\nspace is not well pruned using the closure downward property.\nThus, the utility measure is neither monotone nor antimonotone.To address this limitation the TWU: Transaction\nWeighted Utility property is defined to prune the high utility\npattern space [35, 36]. It is an upper-bound monotone measure\nto reduce the search space. More efficient HUIM algorithms\nbased on TWU have been recently proposed such as EFIM:\nEFficient high-utility Itemset Mining [37], and d[2]HUP: Direct\nDiscovery for High Utility Patterns [38]. The particularity of\nsuch approaches is that they use more efficient data structures\nto determine the TWU and the utility values. Singh et al. [39]\naddress the problem of the minimum utility threshold tuning\nand derived the top k high utility patterns. It uses transaction\nmerging and data projection techniques to reduce the data\nscanning cost. It also develops an intelligent strategy designed\nfor top k high utility patterns to prune the enumeration search\ntree. Gan et al. [40] proposed a correlated high utility pattern\nminer algorithm. It considers the positive correlation, the profitable value concepts, and several strategies to prune the\nsearch space. Lee et al. [41] developed an efficient incremental approach for identifying high utility patterns. It adopts an\naccurate data structure to mine high utility patterns in an incremental way.\n\n\nSPM Sequential Pattern Mining is an extension of FIM to\ndiscover a set of ordered patterns in a sequence database\n\n[42–44]. Salvemini et al. [42] find the completeset of the sequence patterns by reducing the candidates generation runtime\nby employing an efficient lexicographic tree structure.\nFumarola et al. [43] discover closed sequential patterns using\ntwo main steps, i) Finding the closedof sequence patterns of\nsize 1, and ii) Generating new sequences from the sequence\npatterns of size 1, already deduced in the first step. Van et al.\n\n[44] introduced the pattern-growth algorithm in solving the\nsequential pattern mining problem with itemset constraints.\nIt proposed an incremental strategy to prune the enumeration\nsearch tree which allows to reduce the number of visited\nnodes. Aisal et al. [45] proposed a novel convoy pattern mining approach which can operateon a variety of operational\ndata stores. It suggested a new heuristic to prune the objects\nwhich have no chance of forming a convoy. Wu et al. [46]\nsolved the contrast sequential pattern mining problem, which\nis an extension of SPM, discovered all relevant patterns that\nfigure out in one sequence data and not in the others. These\npatterns are highly used in some specified application such as\nanalysing anomalous customers in the business intelligence or\nmedical diagnosis in the smart healthcare [47–49].\n\nHigh performance computing Regarding high performance\ncomputing, many algorithms have been developed for\nboosting the FIM performance [15, 50–54]. However, few\nalgorithms have been proposed for the other pattern mining\nproblem [16–18, 55]. In [52], some challenges in big data\nanalytics are discussed, such as mining evolving data streams\nand the need to handle many exabytes of data across various\napplication areas such as social network analysis. The\nBigFIM: Big Frequent Itemset Mining [56] algorithm is presented, which combines principles from both Apriori and\nEclat. BigFIM is implemented using the MapReduce paradigm. The mappers are determined using Eclat algorithm,\nwhereas, the reducers are computed using the Apriori algorithm. [57] develops two strategies for parallelizing both candidate itemsets generation and support counting on a GPU\n(Graphic Processing Unit). In the candidate generation, each\nthread is assigned two frequent (k-1)-sized itemsets, it compares them to make sure that they share the common (k-2)\nprefix and then generates a k-sized candidate itemset. In the\nevaluation, each thread is assigned one candidate itemset and\ncounts its support by scanning the transactions simultaneously. The evaluation of frequent itemsets is improved in [58] by\nproposing mapping and sum reduction techniques to merge all\ncounts ofthe given itemsets. It is also improved in [59] by\ndeveloping three strategies for minimizing the impact of the\ngraphical processor unit thread divergence. In [60], a multilevel layer data structure is proposed to enhance the support\ncounting of the frequent itemsets. It divides vertical data into\nseveral layers, where each layer is an index table of the next\n\n\n-----\n\nlayer. This strategy can completely represent the original vertical structure. In a vertical structure, each item corresponds to\na fixed-length binary vector. However, in this strategy, the\nlength of each vector varies, which depends on the number\nof transactions included in the corresponding item. A Hadoop\nimplementation based on MapReduce programming approach\ncalled FiDoop: Frequent itemset based on Decomposition is\nproposed in [61] for the frequent itemsets mining problem. It\nincorporates the concept of FIU-tree (Frequent Itemset\nUltrametric tree) rather than traditional FP-tree of FPGrowth algorithm, for the purpose of improving the storage\nof the candidate itemsets. An improved version called\nFiDoop-DP is proposed in [15]. It develops an efficient strategy to partition data sets among the mappers. This allows\nbetter exploration of cluster hardware architecture by avoiding\njobs redundancy. Andrzejewski et al. [62] introduce the concept of incremental co-location patterns, i.e., update the set of\nknowledge about the spatial features after inserting new spatial data to the original one. The authors develop a new parallel\nalgorithm which combines effective update strategy and multi\nGPU co-location pattern mining [63] by designing an efficient\nenumeration tree on GPU. Since the proposed approach is\nmemory-aware, i.e., the data is divided into several package\nto fit to GPU memories, it only achieves an speedup of six.\nJiang et al. [64] adopt a parallel FP-Growth for mining world\nocean atlas data. The whole data is partitioned among multiple\nCPU threads, where each thread explores 300,000 data points\nand derives correlation and regularities of oxygen, temperature, phosphate, nitrate, and silicate in the ocean. The experimental results reveal that the suggested adaptation only\nreachesan speedup of 1.2. Vanhalli et al. [65] developed a\nparallel row enumerated algorithm for mining frequent colossal closed patterns from high dimensional data. It first prunes\nthe whole data by removing irrelevant items and transactions\nusing rowset cardinality table, which determines the closeness\nof each subset of transactions. In addition, it uses a hybrid\nparallel bottom-up bitset based approach to enumerate the\ncolossal frequent closed patterns. This approach is fast,\nhoweverit suffers from the accuracy issue, which may ignore\nsome relevant patterns due to the preprocessing phase. It also\nrequires additional parameter to be fixed, represented by a\ncardinality threshold. Yu et al. [66] propose the parallel version of PrefixSpan on Spark: PrefixSpan-S. It optimizes the\noverhead by first loading the data from the Hadoop distributed\nfile system into the RDDs: Resilient Distributed Datasets, and\nthen reading the data from RDDs, and save the potential results back into the RDDs. This approach reaches a good performance with a wise choice of the minimum support threshold. However, it is very sensitive to the data distribution.\nKuang et al. [67] proposed the parallel implementation of\nFP-Growth algorithm in Hadoop by removing the data redundancy between the different data partitions, which allows to\nhandle the transactions in a single pass. Sumalatha et al. [68]\n\n\nintroduces the concept of distributed temporal high utility sequential patterns, and propose an intelligent strategy by creating a time interval utility data structure for evaluating the\ncandidate patterns. The authors also defined two utility upper\nbounds, remaining utility, and co-occurrence utility to prune\nthe search space.\nTo improve the runtime performance of the pattern mining\napproaches, several strategies have been proposed using\nmetaheuristics, specifically exploiting evolutionary and/or\nswarm intelligence approaches [13, 14, 69, 70]. However,\nthese optimizations are inefficient when dealing with large\nand big transactional databases where only few number of\ninteresting patterns are discovered. To deal with this challenging issue, the next section presents a new framework, which\ninvestigates both decomposition techniques and distributed\ncomputing in solving pattern mining algorithms.\n\n### 4 DT-DPM: decomposition transaction for distributed pattern mining\n\nThis section presents the DT-DPM (Decomposition Transaction\nfor Distributed Pattern Mining) framework, that integrates the\nDBSCAN: Density-Based Spatial Clustering of Applications\nwith Noise algorithm and distributed computing represented\nby MapReduce, CPU multi-cores and Single CPU for solving\npattern mining problems. As seen in Fig. 1, the DT-DPM\nframework uses heterogeneous distributed computing and decomposition techniques for solving pattern mining problems.\nDetailed explanation of the DT-DPM framework, step by step,\nis given in the following.\n\n#### 4.1 DBSCAN\n\nThe aim of this step is to divide a database into a collection of\nhomogeneous groups using decomposition techniques, where\neach group shares entries highly correlated, i.e., the database\nentries of each group share maximum number of items compared to the entries of the other groups.\n\nDefinition 13 A database is decomposed into several groups\nD\nG = {Gi}, where each group Gi is subset of rows in D such as\nGi ∪ Gj = ∅. We define, I Gð iÞ, the set of items of the group Gi\nby\n\nI Gð iÞ ¼ �⋃I D� j�=D j∈Gi� ð11Þ\n\nProposition 1 Suppose that the groups in G share any items\nwhich means\n\n∀�Gi; G j�∈G[2]; ∨i≠ j; I Gð iÞ∩I G� j� ¼ ∅ ð12Þ\n\n\n-----\n\nFig. 1 DT-DPM framework\n\nWe have the following proposition\n\n� k �\n\nL ⋃ 13\n¼ i¼1 [L][i] ð Þ\n\nNote that Pi is the set of the relevant patterns of the group\nGi.\nFrom the above proposition, one may argue that if the\nwhole transactions in the database are decomposed in such a\nway, the independent groups will be derived. It means that,\nany group of transactions share items with any other group,\nand therefore, the groups could be solved separately.\nUnfortunately, such case is difficult to realize, as many dependencies may be observed between rows. The aim of the decomposition techniques is to minimize the separator items\nbetween the groups such as\n\nG[*] ¼ Garg min j⋃�I Gð iÞ∩I G� j��j ð14Þ\n\nThe aim of the decomposition step is to minimize the\nshared items between the different clusters, these shared items\nare called Separator Items. More formally, this decomposition\ngenerates a labeled non-directed graph weighted noted G = <\nC, S> where C is the set of nodes formed by the clusters of G\nand S is the set of the separator items. Each element s[ij] in S\ncontains two components: i) s[ij]l [, is the label of the element][ s][ij][, it]\nis represented by the set of items shared by the clusters Ci and\nCj. ii)s[ij]w [is the weight of the element][ s][ij][, it is represented by the]\nnumber of items shared by the cluster Ci and Cj. As a result, k\n\n\ndisjoint partitions P = {P1, P2, …, Pk}, where Pi ∩ Pj = ∅,\n∀(i, j) ∈ [1, …, k][2] and ⋃[k]i¼1[P][i][ ¼][ T] [. The partitions are con-]\nstructed by minimizing the following function\n\n\n∑k ∑k j∑Pij �2[G]i[l] −1�−∑jP jj �2[G][l]j−1�!!\n\ni¼1 j¼1 l¼1 l¼1\n\n\n15\nð Þ\n\n\nSolving this equation by exact solver requires high\ncomputational time. One way to solve this issue is to\nuse clustering algorithms [71]. The adaptation of the\nDBSCAN [72] clustering algorithms has been investigated in this research. Before this, some definitions are\ngiven as follows.\n\nDefinition 14 (transaction-based similarity) We define\nTransaction-based similarity by an adopted Jaccard similarity\n\n[73] as\n\n\nJ D D� iD j� ¼\n\n\n∑x∈ðDi∩D jÞ[Sim][ D]� [i][;][ D] [j][;][ x]� 16\n\njDij þ jD jj þ ∑x∈ðDi∩D jÞ[Sim]�[ D][i][;][ D] [j][;][ x]� ð Þ\n\n\n� 1 if x[i] x [j]\n\nSim D� i; D j; x� ¼ 0 otherwise¼ ð17Þ\n\nNote that x[i] is the value of the variable x in the row data Di.\n\nDefinition 15 (centroids) We define the centroids of the group\nof rows data Gi, noted Gi by\n\n\n-----\n\n� �\nGi ¼ ⋃ maxxl ��x[i]l jxl∈X Gð iÞ ð18Þ\n\nwhere max ��x[i]l is the most value of the variable xl in the group\nxl\n\nGi.\n\nDefinition 16 (clause neighborhoods) We define the neighborhoods of a row data Di for a given threshold ϵ, noted N Di by\n\nN Di ¼ �D jjJ D D� iD j�≥ ϵ∨ j≠i� ð19Þ\n\nDefinition 17 (core data) A row data Di is called core data if\nthere is at least the minimum number of rows data σD such as\njN Dij≥ σD\nAlgorithm 1 presents the pseudo-code of the decomposition for the rows data. The process starts by checking the ϵneighborhood of each transaction. The core transactions are\ndetermined and then iteratively collects density-reachable\ntransactions from these core transactions directly, which may\ninvolve merging a few density-reachable clusters. The process\nterminates when no new transactions can be added to any\ncluster. The output of the decomposition step is a labeled\nnon-directed graph weighted noted O = < G, S> where G is\nthe set of nodes formed by the groups of the rows data and S is\nthe set of the separator items. Each element s[ij] in S contains\ntwo components: i) s[ij]l [, is the label of the element][ s][ij][, it is]\n\n\nrepresented by the set of variables by the groups Gi and Gj.\nii)s[ij]w [is the weight of the element][ s][ij][, it is represented by the]\nnumber of variables shared by the group Gi and Gj.\n\n#### 4.2 Mining process\n\nAfter applying the DBSCAN algorithm on the whole data, the\nnext step is to solve each resulted cluster independently using\n\nheterogeneous machines such as supercomputers, CPU cores,\nand single CPU core. The question that should be answered\nnow is, which cluster should be assigned to which machine?\nWe know that supercomputers are most powerful than CPU\ncores and single CPU. Obviously, clusters with high workload\nare assigned to supercomputer, the remaining clusters are\nassigned to CPU cores, and the noise data are assigned to a\nsingle CPU. This raises the following proposition,\n\nProposition 2 Consider a function Cost(, m, n), that comA\nputes the complexity cost of the given pattern mining algorithm . Consider the output of the DBSCAN algorithm O =\nA\n< G, S>, the mining process by of the clusters could be\nA\nperformed in parallel by using heterogeneous machines according to the density of each cluster Gi represented by\nðjGij; I Gð iÞÞ, and with a complexity cost threshold μcost as\n\n1. If Cost (A, ∣Gi∣, I Gð iÞ ) ≥ μcost then send Gi to\nMapReduce framework.\n\n\n-----\n\n2. If Cost (A, ∣Gi∣, I Gð iÞ ) < μcost then send Gi to CPU\ncores.\n\nMap and reduce Each mapper Mi first recuperates the\nassigned cluster Gi. It then processes the transactions of the\ncluster Gi, and applies the mining process for each transaction.\nGradually, it creates a set of candidate patterns Ci. When Mi\nscans all transactions of the cluster Gi, it sends Ci to the reducer Ri. The reducer Ri scans the candidate patterns Ci, and\ncomputes the local support of each pattern belong to Ci. This\nallows to create the local hash table LHMRi.\n\nCPU cores The CPU cores solve clusters having complexity\ncost less than μcost, where each CPU core deal with one cluster\nof transactions, where it applies sequentially the given pattern\nmining algorithm on the assigned cluster. The result is\nA\nstored in the local hash table called LHCPU i .\n\nSingle CPU The noise transactions resulted by applying\nDBSCAN are solved separately in a single CPU. The generated noises are merged into a small transactions, where the\nsequential process is applied on it. Again, as result, a local\nhash table called LHnoise is stored\n\nMerging The merging step is then performed to determine the\nglobal support of all patterns and extract all relevant patterns\nfrom the global hash table . This step considers the set of\nGH\nseparator items S as wellas the clusters in the mining process.\nThis allows to discover all relevant patterns from the whole\ntransactions. The possible candidate patterns are then generated from the separator items. For each generated pattern, the\naggregation function (see Definition 2) is then used to determine the interestingness of this pattern in the whole transactional database. Note that, the interestingness depends on the\nproblem. For instance, if we are interested to deal with a frequent itemset mining problem, the interestingness function\nshould be the support measure. The relevant patterns of the\nseparator items are then concatenated with the local hash tables to derive the global relevant patterns of the whole\nLH\ntransactional database, noted .\nGH\n\nDefinition 18 Let us define an aggregation function of the\npattern p in the clusters of the transactions C by\n\nk\nAð Þ ¼ p ∑ Interestingness Gð i; I Gð iÞ; pÞ ð20Þ\n\ni¼1\n\nDT-DPM improves the sequential version of the pattern\nmining algorithms by exploiting heterogeneous machines\nwhile generating and evaluating the candidate patterns. The\nload balancing is automatically conducted since the transactions are assigned to the workers using decomposition step.\n\n\nWorkers including (mappers, CPU cores, and single CPU) can\nprocess transactions independently and stored the results into\nthe local hash tables. The merging step is finally performed on\nthe local hash tables and extract and extract the set of all\nfrequent patterns by dealing with the separator items using\nthe aggregation function.\n\n### 5 Performance evaluation\n\nExtensive experiments were carried out to evaluate the DTDPM framework. Five case studies were investigated by considering the FIM, WIM, UIM, HUIM, and SPM problems.\nDT-DPM is integrated on the SPMF data mining library\n\n[74]. It offersmore than 150 pattern mining algorithms. DTDPM java source code is integrated on the five best pattern\nmining algorithms in terms of time complexity [8]: i) frequent\nitemset mining: PrePost+ [75], ii) weighteditemset mining:\nWFIM [27], iii) uncertain high utility itemset mining: UApriori [76], iv) high utility itemset mining: EFIM [37], and\nv) sequential pattern mining: FAST [42]. All experiments\nwere run on a computer with an Intel core i7 processor running\nWindows 10 and 16 GB of RAM.\n\n#### 5.1 Dataset description\n\nExperiments were run on well-known pattern mining large\nand big databases. Table 1 presents the characteristics of the\nlarge databases used in the experiments. Moreover, three very\nlarge databases have been used to evaluate the patternmining\napproaches:\n\n1. Webdocs[1] It is created from a set of links among html\ndocuments available on the Web. It contains 1,692,082\ntransactions with 526,765 different items. The maximum\nnumber of items per transactions of this database is\n70,000 with 1.48 GB as database size [77].\n2. Twitter[2] It is dataset related to user tweets. It contains\n41.7 million user profiles, 1.47 billion social relations,\n4,262 trending topics, and 106 million tweets [78].\n3. NewYorkTimes[3] It consists of over 1.8 million newspaper articles published among twenty years.\n\nIn addition, an IBM Synthetic Data Generator for Itemsets\nand Sequences[4] is used to generate big databases of different\nnumber of items and different number of transactions.\n\n[1 http://fimi.ua.ac.be/data/webdocs](http://fimi.ua.ac.be/data/webdocs)\n[2 http://an.kaist.ac.kr/traces/WWW2010.html](http://an.kaist.ac.kr/traces/WWW2010.html)\n[3 http://corpus.nytimes.com/](http://corpus.nytimes.com/)\n[4 https://github.com/zakimjz/IBMGenerator](https://github.com/zakimjz/IBMGenerator)\n\n\n-----\n\nTable 1 Large databases\nProblem Data set Data set ∣D∣ ∣I∣ Avg. Size/\nType Name Avg. Seq. Size\n\nDense Accident 340,183 468 33.8\nDense Chess 3196 75 37.0\nDense Connect 67,557 129 43.0\nFIM, WIM, Dense Mushroom 8124 119 23.0\nUIM, HUIM Dense Pumsb 49,046 2113 74.0\nVery Sparse Korasak 990,000 41,270 8.1\nSparse Foodmart 4141 1559 4.4\nVery Sparse Chainstore 1,112,949 46,086 7.2\nBook Leviathan 5834 9025 33.81\nSPM Language Sign 730 267 51.99\n\n\nProtein Snack 163 20 60\nWeb Stream FIFA 20,450 2990 34.74\n\n\n#### 5.2 Decomposition performance\n\nThe first experiment aims at evaluating the clustering step.\nSeveral tests have been performed on the k-means algorithm\nto fix the number of clusters k. Figures 2 and 3 present both the\nquality of obtained clusters measures by the percentage of\nseparator items and the runtime performance in seconds using\nthe twelve databases mentioned above. By varying with the\nnumber of clusters from 2 to 20, the percentage of the separator items is reduced. Thus, when tk is set to 2, the percentage\nof the separator items exceeds 50%, while this percentage\ndoes not reach 1% when setting the number of clusters is set\nto 20 for almost databases. Moreover, by increasing the number of clusters, the runtimeincreases linearly, where it does not\nreach 2.50 seconds for all databases. As a result, we fix the\nnumber of clusters to 20 for the remaining of the experimentation. Given such results we can conclude that the preprocessing step does not influence on the overall performance\nof the DT-DPM framework, in particular for very large and big\ndatabases.\nFigure 3 presents the runtime performance of the pattern\nmining algorithms with and without the DT-DPM framework\nfor both strategies (approximate and exact) using different databases and with different mining thresholds. Experimental\nresults reveal that by reducing the mining threshold and increasing the complexity of the problem solved, the pattern\nmining algorithms benefits from the DT-DPM framework.\nTherefore, for a small value of mining threshold and for more\ncomplex problem like UIM, HUIM or SPM, the approximatebased and exact-based strategies outperform the original pattern mining algorithms. For instance, when the minimum utility threshold is set to 1600 K, the runtime of original EFIM and\nEFIM using the DT-DPM framework is 1 s on Connect database. However, when setting the minimum utility to 1000 K,\nthe runtime of original EFIM exceeds 8,000 seconds, and the\nruntime of EFIM with DT-DPM framework does not reach\n\n|Col1|Col2|Col3|\n|---|---|---|\n||kmeans DBSCAN Intuitive \u0001 \u0001 \u0001\u0001 \u0001 \u0001 \u0001 \u0001 \u0001 \u0001 \u0001 \u0001 \u0001 \u0001 \u0001 \u0001 \u0001 \u0001 \u0001\u0001 \u0001 \u0001||\n||||\n|\u0001|||\n|\u0001 \u0001|||\n||||\n\n\nDatabase\n\nFig. 2 Percentage(%) of separator items of the decomposition using\ndifferent clustering algorithms on the following databases: D1: pumsb,\nD2: mushroom, D3: connect, D4: chess, D5: accident, D6: korasak, D7:\nfoodmart, D8: chainstore, D9: leviathan, D10: sign, D11: snack, and D12:\nFIFA\n\n\n1,500 seconds. These results are obtained thanks to many factors: i) the decomposition method applied in the DT-DPM\nframework by minimizing the number of separator items, ii)\nsolving sub-problems with small number of transactions and\nsmall number of items, instead of dealing the whole transactional database with the whole distinct items, and iii) the ability\nof the pattern mining algorithms to be integrated with the DTDPM framework.\n\n#### 5.3 Speedup of DT-DPM\n\n\nTable 2 presents the speedup of the pattern mining algorithms\nwith and without using the DT-DPM framework on the large\ndatabases. The results reveal that by increasing the number of\nmappers and the increasing the complexity of the problem\nsolved, the speedup of the pattern mining algorithms benefits\nfrom the DT-DPM framework. Thus, for a large number of\nmappers, and for a more complex problem like UIM, HUIM\nor SPM, the mining process with DT-DPM outperform the\n\n100\n\n\n90\n\n80\n\n70\n\n60\n\n50\n\n40\n\n30\n\n20\n\n10\n\n0\n\n\nD1 D2 D3 D4 D5 D6 D7 D8 D9 D10 D11 D12\n\n\n-----\n\n100\n\n90\n\n80\n\n70\n\n60\n\n50\n\n40\n\n30\n\n20\n\n10\n\n\n250\n\n200\n\n\n150\n\n100\n\n\n0\n\n300\n\n250\n\n200\n\n150\n\n100\n\n\n50\n\n0\n\n|Col1|Col2|Col3|\n|---|---|---|\n||PrePost+ PrePost+(kmeans) \u0001 \u0001 \u0001 \u0001||\n|\u0001|||\n|\u0001|||\n||\u0001||\n\n|Col1|Col2|Col3|\n|---|---|---|\n||WFIM WFIM(kmeans) \u0001 \u0001||\n|\u0001|||\n||||\n||\u0001 \u0001 \u0001 \u0001 \u0001||\n\n\nD1 D2 D3 D4 D5 D6\n\nDatabase\n\n\nD1 D2 D3 D4 D5 D6 D7 D8\n\nDatabase\n\n\n### FIM WIM\n\n1800\n\nEFIM\n\n� EFIM(kmeans)\n\n� 1500\n\n1200\n\n\n50\n\n0\n\n\n900\n\n600\n\n300\n\n0\n\n|Col1|Col2|Col3|\n|---|---|---|\n||U−Apriori U−Apriori(kmeans) \u0001 \u0001 \u0001 \u0001 \u0001||\n||||\n|\u0001|||\n||\u0001 \u0001||\n\n|Col1|Col2|Col3|\n|---|---|---|\n||EFIM EFIM(kmeans)||\n||||\n|\u0001|||\n||\u0001 \u0001 \u0001 \u0001 \u0001 \u0001 \u0001||\n\n\nD1 D2 D3 D4 D5 D6 D7 D8\n\nDatabase\n\n\nD1 D2 D3 D4 D5 D6 D7 D8\n\nDatabase\n\n\n### UIM HUIM\n\nFAST\n\n350 � FAST(kmeans)\n\n300\n\n250\n\n200\n\n150 � � �\n\n100 �\n\n\n50\n\n0\n\n|Col1|FAST|Col3|\n|---|---|---|\n||\u0001 FAST(kmeans) \u0001 \u0001 \u0001 \u0001||\n||||\n\n\nD1 D2 D3 D4\n\nDatabase\n\n\n### SPM\n\nFig. 3 Runtime (seconds) of pattern mining algorithm without and with decomposition step on the following databases: D1: pumsb, D2: mushroom, D3:\nconnect, D4: chess, D5: accident, D6: korasak, D7: foodmart, D8: chainstore, D9: leviathan, D10: sign, D11:snack, and D12: FIFA\n\n\noriginal pattern mining algorithms. For instance, when the\nnumber of mappers set to 2, the speedup of the original\nPrePost+ and PrePost using the DT-DPM framework is less\nthan 5 in the pumsb database. However, by setting the number\nof mappers to 32, the speedup of the original EFIM does not\nreach 8630, and the speedup of the EFIM with DT-DPM\nframework exceeds 800 on Connect database. These results\nare achieved thanks to the following factors, i) The decomposition method applied to the DT-DPM framework by\n\n\nminimizing the number of theseparator items, and ii)\nSolving the sub-problems with small number of transactions\nand small number of items using the Mapreduce architecture.\n\n#### 5.4 DT-DPM Vs state-of-the-art algorithms\n\n\nFigures 4 presents the speedup of DT-DPM against the baseline pattern mining algorithms (FiDoop-DP [15] for FIM,\nPAWI: Parallel Weighted Itemsets [16] for WIM,\n\n\n-----\n\nTable 2 Speedup of pattern mining algorithms with and without using DT-DPM framework using different mappers (2, 4, 8, 16, 32)\n\nProblem Database Without DT-DPM With DT-DPM\n\nMappers 2 4 8 16 32 2 4 8 16 32\n\npumsb 2 3 7 11 34 5 9 15 19 35\nmushroom 3 8 10 13 36 6 11 19 22 39\nFIM: connect 5 10 12 15 39 10 15 19 26 45\nPrePost+ chess 9 12 14 16 41 13 18 23 29 52\naccident 12 16 20 22 45 19 23 29 36 66\nkorasak 2 2 5 9 11 3 4 6 10 14\npumsb 4 8 11 14 37 9 12 19 22 41\nmushroom 7 10 12 15 39 11 15 23 28 49\nconnect 11 14 18 21 45 19 23 29 35 57\nWIM: chess 13 18 23 29 49 23 31 39 44 62\nWFIM accident 29 35 52 59 68 35 47 63 74 88\nkorasak 2 4 6 8 13 5 8 10 13 19\nfoodmart 3 5 9 11 14 6 9 12 15 22\nchainstore 33 39 55 66 75 36 51 66 79 95\n\npumsb 3 9 12 16 39 11 13 18 23 43\nmushroom 5 8 10 14 33 7 8 15 23 42\nconnect 15 17 21 25 48 23 29 36 45 59\nUIM: chess 16 19 25 33 52 18 26 31 46 72\nWFIM accident 30 39 55 66 75 42 52 69 79 90\nkorasak 3 7 10 12 18 6 12 19 25 32\nfoodmart 4 7 12 16 20 9 14 21 29 38\nchainstore 41 55 67 81 99 57 66 81 98 117\npumsb 51 76 91 105 133 69 91 111 142 163\nmushroom 32 49 60 71 82 36 55 71 89 101\nconnect 161 245 367 415 629 254 338 459 510 801\nHUIM: chess 44 59 80 85 91 49 60 88 96 104\nEFIM accident 79 127 163 198 300 121 201 293 368 411\nkorasak 23 35 44 60 68 28 47 52 73 86\nfoodmart 8 15 27 33 43 15 32 44 53 63\nchainstore 48 72 88 94 100 58 84 106 127 135\nleviathan 19 36 52 82 103 32 44 69 99 125\nSPM: sign 41 78 93 105 119 55 86 118 133 151\n\nFAST snack 39 70 90 99 108 50 78 110 145 173\nFIFA 60 92 117 136 175 90 109 142 178 213\n\nThe values in bold represent the best values obtained during the the experiment\n\n\nMRGrowth: MapReduce for FP-Growth [17] for UIM, PHIMiner: Parallel High utility Itemset Miner [55] for HUIM, and\nMG-FSM: Mind the Gap for Frequent Seauence Mining [18]\nfor SPM) using the big databases, and setting the mining\nthreshold to10% for Webdocs, 5% for Twitter and 2% for\nNewYorkTimes database. The results reveal that by increasing\nthe percentage of transactions from 5% to 100% and increasing the complexity of the problem solved, DT-DPM\noutperformed the baseline pattern mining algorithms and benefits from the intelligent partition of the transactional database\nand the intelligent mapping of the different clusters into the\n\n\nmapper nodes. For instance, the speedup of our framework is\n701 on Webdocs whereas the speedup of FiDoop-DP is 521.\nOn the other hand, for mining the speedup of our framework\non NewYorkTimes is 1325 whereas the speedup of MG-FSM\ndoes not exceed 900.\n\n#### 5.5 Results on big databases\n\nFigures 5 present the runtime of DT-DPM and both baseline\nMapReduce based models using big databases for solving\nboth Itemset and Sequential Pattern Mining problems. The\n\n\n-----\n\n900\n\n800\n\n700\n\n600\n\n500\n\n400\n\n300\n\n200\n\n100\n\n\n### FIM(Webdocs)\n\n5 10 20 50 80 100\n### % Transactions\n\n\n1200\n\n1000\n\n\n800\n\n600\n\n\n400\n\n200\n\n\n0\n\n1400\n\n\n0\n\n\n### WIM(Twitter)\n\n5 10 20 50 80 100\n### % Transactions\n\n\n1200\n\n1000\n\n\n### FIM Problem (Webdocs) WIM Problem (Twitter) UIM(Twitter) HUIM(Twitter)\n\n1400\n\nDT-DPM\nMRGrowth\n\n1200\n\n\n800\n\n600\n\n\n1000\n\n800\n\n\n400\n\n200\n\n\n600\n\n400\n\n\n0\n\n\n200\n\n0\n\n\n5 10 20 50 80 100\n### % Transactions\n\n\n5 10 20 50 80 100\n### % Transactions\n\n\n### UIM Problem (Twitter) HUIM Problem (Twitter)\n\n\n1600\n\n1400\n\n\n1200\n\n1000\n\n\n800\n\n600\n\n\n400\n\n200\n\n\n0\n\n\n### SPM(NewYorkTimes)\n\n5 10 20 50 80 100\n### % Transactions\n\n\n### SPM Problem (NewYorkTimes)\n\nFig. 4 Speedup of DT-DPM and the state-of-the-art parallel pattern mining algorithms using big databases\n\n\n-----\n\n450\n\n400\n\n350\n\n300\n\n250\n\n200\n\n150\n\n100\n\n50\n\n0\n\n\n**# Transactions=1M**\n\n\n**# Transactions=2M**\n\n\n450\n\n400\n\n350\n\n300\n\n250\n\n200\n\n150\n\n100\n\n50\n\n0\n\n|Col1|DT-|DPM|Col4|Col5|Col6|Col7|Col8|Col9|Col10|\n|---|---|---|---|---|---|---|---|---|---|\n|||||||||||\n||FiDo NG-|op-D PFP|P|||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n\n|Col1|DT-D|PM|Col4|Col5|Col6|Col7|Col8|Col9|Col10|\n|---|---|---|---|---|---|---|---|---|---|\n|||||||||||\n||FiDo NG-|op-D PFP|P|||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n\n\n550\n\n500\n\n450\n\n400\n\n350\n\n300\n\n250\n\n200\n\n150\n\n100\n\n50\n\n0\n\n\n0 10 20 30 40 50 60 70 80 90 100\n# Items X 10,000\n\n**# Transactions=5M**\n\n|Col1|DT-|DPM|Col4|Col5|Col6|Col7|Col8|Col9|Col10|\n|---|---|---|---|---|---|---|---|---|---|\n|||||||||||\n||FiDo|op-D|P|||||||\n|||||||||||\n||NG-|PFP||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n\n|Col1|DT-|DPM|Col4|Col5|Col6|Col7|Col8|Col9|Col10|\n|---|---|---|---|---|---|---|---|---|---|\n|||||||||||\n||FiDo|op-D|P|||||||\n|||||||||||\n||NG-|PFP||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n\n\n0 10 20 30 40 50 60 70 80 90 100\n# Items X 10,000\n\n\n0 10 20 30 40 50 60 70 80 90 100\n# Items X 10,000\n\n**# Transactions=10M**\n\n\n0 10 20 30 40 50 60 70 80 90 100\n# Items X 10,000\n\n\n550\n\n500\n\n450\n\n400\n\n350\n\n300\n\n250\n\n200\n\n150\n\n100\n\n50\n\n0\n\n\n## IBM Synthetic Data Generator for Itemsets\n\n\n250\n\n200\n\n\n**# Transactions=1M**\n\n\n**# Transactions=2M**\n\n\n150\n\n100\n\n\n50\n\n0\n\n\n350\n\n300\n\n250\n\n200\n\n150\n\n100\n\n50\n\n0\n\n|Col1|DT-D Prefi|PM xSpa|n-S|Col5|Col6|Col7|Col8|Col9|Col10|\n|---|---|---|---|---|---|---|---|---|---|\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n\n|Col1|Col2|# T|Tran|nsac|ction|ns=2|2M|Col9|Col10|\n|---|---|---|---|---|---|---|---|---|---|\n||DT- Prefi|DPM xSpa|n-S|||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n\n\n350\n\n300\n\n250\n\n200\n\n150\n\n100\n\n50\n\n0\n\n\n0 10 20 30 40 50 60 70 80 90 100\n# Items X 10,000\n\n**# Transactions=5M**\n\n|Col1|DT-|DPM|Col4|Col5|Col6|Col7|Col8|Col9|Col10|\n|---|---|---|---|---|---|---|---|---|---|\n|||||||||||\n||Prefi|xSpa|n-S|||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n\n|Col1|DT- Prefi|DPM xSpa|n-S|Col5|Col6|Col7|Col8|Col9|Col10|\n|---|---|---|---|---|---|---|---|---|---|\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n|||||||||||\n\n\n0 10 20 30 40 50 60 70 80 90 100\n# Items X 10,000\n\n\n0 10 20 30 40 50 60 70 80 90 100\n# Items X 10,000\n\n**# Transactions=10M**\n\n\n0 10 20 30 40 50 60 70 80 90 100\n# Items X 10,000\n\n\n400\n\n350\n\n300\n\n250\n\n200\n\n150\n\n100\n\n50\n\n0\n\n\n## IBM Synthetic Data Generator for Sequences\n\nFig. 5 Comparison of the Runtime of the DT-DPM and the baseline parallel-based pattern mining algorithms using big databases\n\n\n-----\n\nbaseline methods used is FiDoop-DP [15], and NG-PFP:\nNonGroup Parallel Frequent Pattern mining [67] for itemset\nmining, and PrefixSpan-S [66] for sequence mining. The results reveal that our model outperforms the baseline\nMapReduce based models in terms of computational time\nfor both itemset and sequence mining. These results have been\ncarried out whatever the number of items, and the number of\ntransactions. Moreover, when the number of items varied from\n10,000 to 1 million, and the number of transactions from 1\nmillionto 10 million, the ratio of runtimes between our model\nand the baseline models is increased. All these results are\nobtained thanks to, i) the k-means algorithm which minimizes\nthe number of shared items, and ii) the efficient hybrid parallel\nprocessing of MapReduce, and the Multi CPU cores, which\ntakes into account the information extracted during the decomposition step.\n\n### 6 Conclusion\n\nA novel distributed pattern mining framework called DTDPM is proposed in this paper. DT-DPM aims to derive relevant patterns on big databases by studying the correlations\nwithin the transactional database and exploring heterogeneous\narchitectures. The set of transactions are first grouped using a\nclustering approach, where transactions close to each other are\nassigned to the same cluster. For each cluster of transactions,\nthe pattern mining algorithm is launched in order to discover\nthe relevant patterns. DT-DPM solves the issue of cluster’s\nsize by incorporating heterogeneous computing such as single\nCPU, CPU cores, and MapReduce. Thus, the noise transactions are assigned to the single CPU core, the micro clusters\nare assigned to CPU cores, and dense clusters are assigned to\nthe MapReduce reduce architecture. Experimental evaluation\nof DT-DPM was integrated on the SPMF tool, where five case\nstudies have been shown (FIM, WIM, UIM, HUIM and\nSPM). The results reveal that by using DT-DPM, the scalability performance of pattern mining algorithms have been improved on large databases. Moreover, DT-DPM outperforms\nthe baseline pattern mining algorithms on big databases.\nMotivated by the promising results shown in this paper, we\nplan to boost the performance ofDT-DPM and apply it on big\ndata mining applications such as twitter analysis, smart building applications, and other large-scale applications.\n\nFunding Information Open Access funding provided by NTNU\nNorwegian University of Science and Technology (incl St. Olavs\nHospital - Trondheim University Hospital).\n\nOpen Access This article is licensed under a Creative Commons\nAttribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as\nyou give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were\nmade. The images or other third party material in this article are included\n\n\nin the article's Creative Commons licence, unless indicated otherwise in a\ncredit line to the material. If material is not included in the article's\nCreative Commons licence and your intended use is not permitted by\nstatutory regulation or exceeds the permitted use, you will need to obtain\npermission directly from the copyright holder. To view a copy of this\n[licence, visit http://creativecommons.org/licenses/by/4.0/.](http://creativecommons.org/licenses/by/4.0/)\n\n### References\n\n1. H. Belhadi, K. Akli-Astouati, Y. Djenouri, and J. C.-W. Lin Data\nmining-based approach for ontology matching problem. Appl\nIntell, pp. 1–18\n2. Djenouri Y, Belhadi A, Fournier-Viger P (2018) Extracting useful\nknowledge from event logs: a frequent itemset mining approach.\nKnowl-Based Syst 139:132–148\n3. Djenouri Y, Belhadi A, Belkebir R (2018) Bees swarm optimization\nguided by data mining techniques for document information retrieval. Expert Syst Appl 94:126–136\n4. Djenouri Y, Djamel D, Djenoouri Z (2017) Data-mining-based decomposition for solving MAXSAT problem: towards a new approach. IEEE Intell Syst, vol. In press, pp. 1–15\n5. He Z, Zhang S, Gu F, Wu J (2019) Mining conditional discriminative sequential patterns. Inf Sci 478:524–539\n6. Choi H-J, Park CH (2019) Emerging topic detection in twitter\nstream based on high utility pattern mining. Expert Syst Appl\n115:27–36\n7. Djenouri D, Laidi R, Djenouri Y, Balasingham I (2019) Machine\nlearning for smart building applications: Review and taxonomy.\nACM Comput Surv (CSUR) 52(2):24\n8. Fournier-Viger P, Lin JC-W, Vo B, Chi TT, Zhang J, Le HB (2017)\nA survey of itemset mining. Wiley Interdisciplinary Reviews: Data\nMining and Knowledge Discovery (4):7, e1207\n9. Aggarwal CC, Han J (2014) Frequent pattern mining. Springer\n10. Goethals B (2003) Survey on frequent pattern mining. Univ Hels\n19:840–852\n11. Mabroukeh NR, Ezeife CI (2010) A taxonomy of sequential pattern\nmining algorithms. ACM Comput Surv (CSUR) 43(1):3\n12. Hsieh Y-H, Chen C-C, Shuai H-H, Chen M-S (2018) Highly parallel sequential pattern mining on a heterogeneous platform. in\nIEEE International Conference on Data Mining, pp. 1037–1042\n13. Zhang L, Fu G, Cheng F, Qiu J, Su Y (2018) A multi-objective\nevolutionary approach for mining frequent and high utility itemsets.\nAppl Soft Comput 62:974–986\n14. Djenouri Y, Comuzzi M (2017) Combining apriori heuristic and\nbio-inspired algorithms for solving the frequent itemsets mining\nproblem. Inf Sci 420:1–15\n15. Xun Y, Zhang J, Qin X, Zhao X (2017) FiDoop-DP: data\npartitioning in frequent itemset mining on hadoop clusters. IEEE\nTransactions on Parallel and Distributed Systems 28(1):101–114\n16. Baralis E, Cagliero L, Garza P, Grimaudo L (2015) Pawi: Parallel\nweighted itemset mining by means of mapreduce, in IEEE\nInternational Congress on Big Data, pp. 25–32\n17. Leung CK-S, Hayduk Y (2013) Mining frequent patterns from\nuncertain data with mapreduce for big data analytics, in\nInternational Conference on Database Systems for Advanced\nApplications, pp. 440–455\n18. Miliaraki I, Berberich K, Gemulla R, Zoupanos S (2013) Mind the\ngap: Large-scale frequent sequence mining, in ACM SIGMOD\nInternational Conference on Management of Data, pp. 797–808\n19. Djenouri Y, Lin JC-W, Nørvåg K, Ramampiaro H (2019) Highly\nefficient pattern mining based on transaction decomposition, in\n\n\n-----\n\nIEEE International Conference on Data Engineering, pp. 1646–\n1649\n20. Fournier-Viger P, Zhang Y, Lin JC-W, Fujita H, Koh YS (2019)\nMining local and peak high utility itemsets. Inf Sci 481:344–367\n21. Yun U, Kim D, Yoon E, Fujita H (2018) Damped window based\nhigh average utility pattern mining over data streams. Knowl-Based\nSyst 144:188–205\n22. Agrawal R, Imieliński T, Swami A (1993) Mining association rules\nbetween sets of items in large databases. ACM SIGMOD Rec\n22(2):207–216\n23. Han J, Pei J, Yin Y (2000) Mining frequent patterns without candidate generation. ACM SIGMOD Rec 29(2):1–12\n24. Zhao X, Zhang X, Wang P, Chen S, Sun Z (2018) A weighted\nfrequent itemset mining algorithm for intelligent decision in smart\nsystems. IEEE Access 6:29 271–29 282\n25. Yun U, Leggett JJ (2005) WFIM: weighted frequent itemset mining\nwith a weight range and a minimum weight. in SIAM International\nConference on Data Mining, pp. 636–640\n26. Yun U (2007) Efficient mining of weighted interesting patterns with\na strong weight and/or support affinity. Inf Sci 177(17):3477–3499\n27. Yun U (2009) On pushing weight constraints deeply into frequent\nitemset mining. Intelligent Data Analysis 13(2):359–383\n28. Le N-T, Vo B, Nguyen LB, Fujita H, Le B (2019) Mining weighted\nsubgraphs in a single large graph. Inf Sci 514:149–165\n29. Lee G, Yun U, Ryu KH (2017) Mining frequent weighted itemsets\nwithout storing transaction ids and generating candidates.\nInternational Journal of Uncertainty, Fuzziness and KnowledgeBased Systems 25(01):111–144\n30. Li H, Zhang N, Zhu J, Wang Y, Cao H (2018) Probabilistic frequent\nitemset mining over uncertain data streams. Expert Syst Appl 112:\n274–287\n31. Lee G, Yun U, Ryang H (2015) An uncertainty-based approach:\nfrequent itemset mining from uncertain data with different item\nimportance. Knowl-Based Syst 90:239–256\n32. Liaqat M, Khan S, Younis MS, Majid M, Rajpoot K (2019)\nApplying uncertain frequent pattern mining to improve ranking of\nretrieved images. Appl Intell 49(8):2982–3001\n33. Lee G, Yun U (2017) A new efficient approach for mining uncertain\nfrequent patterns using minimum data structure without false positives. Futur Gener Comput Syst 68:89–110\n34. Chan R, Yang Q, and Shen Y-D (2003) Mining high utility itemsets,\nin IEEE International Conference on Data mining, pp. 19–26\n35. Liu Y, Liao W-k, and Choudhary A (2005) A two-phase algorithm for\nfast discovery of high utility itemsets, in Pacific-Asia Conference on\nKnowledge Discovery and Data Mining, pp. 689–695\n36. Lin C-W, Hong T-P, Lu W-H (2011) An effective tree structure for\nmining high utility itemsets. Expert Syst Appl 38(6):7419–7424\n37. Zida S, Fournier-Viger P, Lin JC-W, Wu C-W, Tseng VS (2017)\nEFIM: a fast and memory efficient algorithm for high-utility itemset\nmining. Knowl Inf Syst 51(2):595–625\n38. Liu J, Wang K, and Fung BC (2012) Direct discovery of high utility\nitemsets without candidate generation, in IEEE International\nConference on Data Mining, pp. 984–989\n39. Singh K, Singh SS, Kumar A, Biswas B (2019) TKEH: an efficient\nalgorithm for mining top-k high utility itemsets. Appl Intell 49(3):\n1078–1097\n40. Gan W, Lin JC-W, Chao H-C, Fujita H, Philip SY (2019)\nCorrelated utility-based pattern mining. Inf Sci 504:470–486\n41. Lee J, Yun U, Lee G, Yoon E (2018) Efficient incremental high\nutility pattern mining based on pre-large concept. Eng Appl Artif\nIntell 72:111–123\n42. Salvemini E, Fumarola F, Malerba D, and Han J (2011) Fast sequence mining based on sparse id-lists, in International Symposium\non Methodologies for Intelligent Systems, pp. 316–325\n\n\n43. Fumarola F, Lanotte PF, Ceci M, Malerba D (2016) CloFAST:\nclosed sequential pattern mining using sparse and vertical id-lists.\nKnowl Inf Syst 48(2):429–463\n44. Van T, Vo B, Le B (2018) Mining sequential patterns with itemset\nconstraints. Knowl Inf Syst 57(2):311–330\n45. Orakzai F, Calders T, Pedersen TB (2019) k/2-hop: fast mining of\nconvoy patterns with effective pruning. Proceedings of the VLDB\nEndowment 12(9):948–960\n46. Wu R, Li Q, Chen X (2019) Mining contrast sequential pattern\nbased on subsequence time distribution variation with discreteness\nconstraints. Appl Intell 49(12):4348–4360\n47. Djenouri Y, Belhadi A, Lin J, Cano A (2019) Adapted k nearest\nneighbors for detecting anomalies on spatio-temporal traffic flow.\nIEEE Access 7:10 015–10 027\n48. Belhadi A, Djenouri Y, Lin JC-W, Djenouri D, and Cano A (2020)\nA GPU-based two phase algorithm for identifying taxi frauds, IEEE\nAccess, vol. In Press, pp. 1–14\n49. Belhadi A, Djenouri Y, Lin JC-W, Zhang CC, Cano A (2020)\nExploring pattern mining algorithms for hashtag retrieval problem.\nIEEE Access 8:10 569–10 583\n50. Han E-H, Karypis G, Kumar V (2000) Scalable parallel data mining\nfor association rules. IEEE Trans Knowl Data Eng 12(3):337–352\n51. Zaki MJ (1999) Parallel and distributed association mining: a survey. IEEE Concurr 7(4):14–25\n52. Wu X, Zhu X, Wu G-Q, Ding W (2014) Data mining with big data.\nIEEE Trans Knowl Data Eng 26(1):97–107\n53. Cano A (2018) A survey on graphic processing unit computing for\nlarge-scale data mining. Wiley Interdisciplinary Reviews: Data\nMining and Knowledge Discovery 8(1):e1232\n54. Djenouri Y, Djenouri D, Belhadi A, Cano A (2019) Exploiting GPU\nand cluster parallelism in single scan frequent Itemset mining. Inf\nSci 496:363–377\n55. Chen Y, An A (2016) Approximate parallel high utility itemset\nmining. Big data research 6:26–42\n56. Moens S, Aksehirli E, and Goethals B (2013) Frequent itemset\nmining for big data, in IEEE International Conference on Big\nData, pp. 111–118\n57. Jian L, Wang C, Liu Y, Liang S, Yi W, Shi Y (2013) Parallel data\nmining techniques on graphics processing unit with compute unified device architecture (CUDA). J Supercomput 64(3):942–967\n58. Djenouri Y, Bendjoudi A, Mehdi M, Nouali-Taboudjemat N,\nHabbas Z (2015) GPU-based bees swarm optimization for association rules mining. J Supercomput 71(4):1318–1344\n59. Djenouri Y, Bendjoudi A, Habbas Z, Mehdi M, Djenouri D (2017)\nReducing thread divergence in gpu-based bees swarm optimization\napplied to association rule mining. Concurrency and Computation:\nPractice and Experience 29(9)\n60. Li Y, Xu J, Yuan Y-H, and Chen L (2017) A new closed frequent\nitemset mining algorithm based on GPU and improved vertical\nstructure. Concurrency and Computation: Practice and\nExperience, vol. 29, no. 6\n61. Xun Y, Zhang J, Qin X (2016) FiDoop: parallel mining of frequent\nitemsets using mapreduce. IEEE Transactions on Systems, Man,\nand Cybernetics: systems 46(3):313–325\n62. Andrzejewski W, Boinski P (2019) Parallel approach to incremental\nco-location pattern mining. Inf Sci 496:485–505\n63. Andrzejewski W, Boinski P (2018) Efficient spatial co-location\npattern mining on multiple GPUs. Expert Syst Appl 93:465–483\n64. Jiang Y, Zhao M, Hu C, He L, Bai H, Wang J (2019) A parallel FPgrowth algorithm on World Ocean Atlas data with multi-core CPU.\nJ Supercomput 75(2):732–745\n65. Vanahalli MK, Patil N (2019) An efficient parallel row enumerated\nalgorithm for mining frequent colossal closed itemsets from high\ndimensional datasets. Inf Sci 496:343–362\n66. Yu X, Li Q, Liu J (2019) Scalable and parallel sequential pattern\nmining using spark. World Wide Web 22(1):295–324\n\n\n-----\n\n67. Kuang Z-j, Zhou H, Zhou J-p, Yang K et al (2019) A non-group\nparallel frequent pattern mining algorithm based on conditional\npatterns. Frontiers of Information Technology & Electronic\nEngineering 20(9):1234–1245\n68. Sumalatha S, Subramanyam R (2020) Distributed mining of high\nutility time interval sequential patterns using mapreduce approach.\nExpert Syst Appl 141:112967\n69. Djenouri Y, Djenouri D, Belhadi A, Fournier-Viger P, Lin JC-W,\nBendjoudi A (2019) Exploiting GPU parallelism in improving bees\nswarm optimization for mining big transactional databases. Inf Sci\n496:326–342\n70. Djenouri Y, Djenouri D, Belhadi A, Lin JC-W, Bendjoudi A, and\nFournier-Viger P (2019) A novel parallel framework for\nmetaheuristic-based frequent itemset mining, in IEEE Congress\non Evolutionary Computation, pp. 1439–1445\n71. Jain AK, Murty MN, Flynn PJ (1999) Data clustering: a review.\nACM Comput Surv (CSUR) 31(3):264–323\n72. Ester M, Kriegel H-P, Sander J, Xu X et al. (1996) A density-based\nalgorithm for discovering clusters in large spatial databases with\nnoise, in International Conference on Knowledge Discovery and\nData Mining, vol. 96, pp. 226–231\n\n\n73. Seifoddini H, Djassemi M (1991) The production data-based similarity coefficient versus jaccard’s similarity coefficient. Comput\nInd Eng 21(1–4):263–266\n74. Fournier-Viger P, Gomariz A, Gueniche T, Soltani A, Wu C-W,\nTseng VS (2014) SPMF: a Java open-source pattern mining library.\nJ Mach Learn Res 15(1):3389–3393\n75. Deng Z-H, Lv S-L (2015) PrePost+: An efficient N-lists-based algorithm for mining frequent itemsets via children–parent equivalence pruning. Expert Syst Appl 42(13):5424–5432\n76. Chui C-K, Kao B, and Hung E (2007) Mining frequent itemsets\nfrom uncertain data,” in Pacific-Asia Conference on Knowledge\nDiscovery and Data Mining, pp. 47–58\n77. Lucchese C, Orlando S, Perego R, and Silvestri F (2004) WebDocs:\na real-life huge transactional dataset, in Frequent Itemset Mining\nImplementations, vol. 126\n78. Kwak H, Lee C, Park H, and S. Moon (2010) What is Twitter, a\nsocial network or a news media? in International Conference on\nWorld Wide Web, pp. 591–600\n\nPublisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.\n\n\n-----\n\n"

ERROR: type should be string, got "https://doi.org/10.1007/s40747 021 00466 y\n\n\n**ORIGINAL ARTICLE**\n\n\n# Blockchain-based green big data visualization: BGbV\n\n**Iqra Shahzad[1]** **· Ayesha Maqbool[2]** **· Tauseef Rana[1]** **· Alina Mirza[3]** **· Wazir Zada Khan[4]** **· Sung Won Kim[5]** **·**\n**Yousaf Bin Zikria[5]** **· Sadia Din[5]**\n\nReceived: 29 March 2021 / Accepted: 3 July 2021 / Published online: 10 August 2021\n© The Author(s) 2021\n\n**Abstract**\nThe progression of Internet of Things (IoT) has resulted in generation of huge amount of data. Effective handling and analysis\nof such big volumes of data proposes a crucial challenge. Existing cloud-based frameworks of Big Data visualization are\nrising costs for servers, equipment, and energy consumption. There is a need for a green solution targeting lesser cost and\nenergy consumption with tamper-proof record-keeping, storage, and interactive visualization with only demanded data. We\nhave proposed a Blockchain-based Green big data Visualization (BGbV) solution using Hyperledger Sawtooth for optimum\nutilization of organization resources. BGbV will support current distributed data visualization platforms and guarantee benefits\nlike security and data availability with lesser storage costs. It helps reduce costs by utilizing small resources that are already\navailable and consume less energy, making it an environmentally friendly solution.\n\n**Keywords IoT** Decentralized visualization Blockchain Big data visualization Hyperledger Sawtooth Secure distributed\n - - - - storage Green solution\n \n### Introduction\n\n\n### B Yousaf Bin Zikria\nyousafbinzikria@ynu.ac.kr\n### B Sadia Din\nsaadia.deen@gmail.com\n\nIqra Shahzad\nishahzad.mscs24@students.mcs.edu.pk\n\nAyesha Maqbool\nayesha.maqbool@mcs.edu.pk\n\nTauseef Rana\ntauseefrana@mcs.edu.pk\n\nAlina Mirza\nalinamirza@mcs.edu.pk\n\nWazir Zada Khan\nwazir.zada@cust.edu.pk\n\nSung Won Kim\nswon@yu.ac.kr\n\n1 Department of Computer Software Engineering, MCS,\nNational University of Sciences and Technology (NUST),\nIslamabad, Pakistan\n\n2 Department of Computer Science, NBC, National University\nof Sciences and Technology (NUST), Islamabad, Pakistan\n\n3 Department of Electrical Engineering, MCS, National\nUniversity of Sciences and Technology (NUST), Islamabad,\nPakistan\n\n\nWith the growth of technology, IoT generated Big data and\nits associated analytic have become common in the computing industry. The cloud-based Big data solutions have\nresulted in a significant increase in energy consumption at\nhost data centers. As most of the time, organization’s small\nyet distributed storage resources remain underutilized. A\nconsiderable amount of energy can be conserved by optimizing the utilization of local storage efficiently.\nNowadays, everything is digitally recorded; IoT systems\ngenerate and process massive amounts of data every day.\nHowever, to process and analyze these data are not enough.\nOur brain tends to find patterns more efficiently when the\ndata are visually represented. Data visualization and analytics have an essential role in decision-making for Big data [3].\nData visualization leads to new opportunities by representing the data in innovative and visual means [26]. However,\nwith the huge volume of data generated, interactive, scalable\nvisualization and data processing are considered a constant\n\n4 Department of Computer Science, Capital University of\nScience and Technology, Islamabad, Pakistan\n\n5 Department of Information and Communication Engineering,\nYeungnam University, Gyeongsan 38541, South Korea\n\n## 1 3\n\n\n-----\n\nchallenge. The size of this information, while visualization is\nso excessive that an ordinary processing technique is insufficient. Construction of each perception may requires in one\nhand traversing the whole data or on other reviewing a small\nsubset of the entire data [11]. For this purpose, expensive\nservers are used. Therefore resources and their efficient utilization is a major concern.\nModern IoT systems involve complex interaction among different entities, demanding a secure, efficient, and distributed\ninfrastructure. Therefore, distributed data visualization is\nconsidered an excellent solution to achieve perception and\noverview of this multi-dimensional massive data [19]. Big\ndata visualization is different from the conventional visualization of data. The visualization of Big data involves a\nlarge variety of data. Consequently, high-level data processing techniques and speed are needed to visualize this vast\nvolume of data [22]. There exist multiple web and user-based\ntechniques for the visualization of Big data. These techniques\nare useful and provide decentralization but with an increase\nin cost for buying servers and equipment, which is a significant concern.\nBlockchain techniques are attractive and have gained\nresearchers’ interest due to their decentralized architecture,\nimmutability, and anonymous record-keeping without centralized authority [6]. Blockchains initially gathered interest\nto support cryptocurrency. However, due to its decentralized\nnature, blockchain has emerged as a distributive computing\ntechnology that provides a secure environment to support\nvarious interactions [21].\nBlockchain is considered a highly immutable, decentralized, and modular platform to transfer assets during the last\nfew years. It is considered a generalized approach for implementing decentralized computing applications. Blockchain\ntechnology is a decentralized and unchangeable database\nthat is highly transparent and secure. It uses a peer-to-peer\nnetwork with no central authority. It contains transaction\nrecords, and these records are shared among all participating nodes. These properties of blockchain make it a\nsuitable candidate for decentralized Big data visualization\nwith advantage of achieve cost-effectiveness, high security,\nand readily available results.\nWe have proposed a green solution BGbV as a support to\nexisting distributed visualization systems. The BGbV framework is based on the blockchain platform, i.e., “Hyperledger\nSawtooth” and is targeting to achieve the following features:\n\n– Tamper proof record-keeping of data for distributed data\nvisualization.\n– Utilization of relativity smaller available distributed\nresources.\n– Green solution with optimal and economical resource\nutilization with least energy consumption.\n\n## 1 3\n\n\n– No central point of failure as multiple copies of data are\nkept.\n– Quality enforcement mechanism by ensuring high availability from the most rated node.\n\nFor our research, a big temporal data set for the crime rate\nhas been taken as a case study to perform the proposed\nresearch. Data are gathered from decentralized nodes and\nvisualization is created according to user’s requirements. We\nhave proposed a solution BGbV that will benefit in many\nways, including less cost of buying resources for decentralized computing and lower energy consumption. It will allow\nusers/people and organizations to gain secure and readily\navailable data visualization that is decentralized and provides\nproductive outputs for analysis.\nIn the rest of the paper, “Literature survey” presents the\nliterature review. “BGbV: blockchain-based green big data\nvisualization” section presents the BGbV paradigm. “Analysis of proposed system” discusses features and analysis of\nthe BGbV framework. “Implementation and testing results”\npresents the implementation and testing details of our solution. We conclude this paper in “Conclusions and future\nwork” and suggest future work.\n\n### Literature survey\n\nMany researchers have addressed Big data and its issues\n\n[24,31]. Exploring Big data offers numerous appealing features, but specialists and experts are also confronted with\nmany difficulties while investigating such mines of data [23].\nMany data sets are too large and complicated to manage on\navailable memory units and are distributed across the clusters\nof computers [17]. Due to the growing nature of Big data, it is\nhard to avoid its challenges. The most immediate issues that\nneed to be addressed are storage issues, management issues,\nand processing issues [16]. It is challenging to handle data\nconnectivity issues, storage limitations, and data processing capabilities in real-time Big data [23]. The exponential\ngrowth in structured and unstructured data has compelled the\nneed for an efficient and reliable storage approach. Therefore,\nthe reliability of devices matters a lot concerning storage\napproaches chosen for handling Big data. While processing,\nbackup, and archiving Big data, there could be many challenges like storage medium, data replication, and duplication\n\n[1]. Big data applications use high-speed servers and equipment, which results in increased cost.\nA distributed paradigm is considered a suitable replacement for costly supercomputers. Distributed approaches are\ndecentralized and aim to disrupt the existing central and conventional ways to deal with huge volumes of data. It also\nensures to handle and deal with new expanding client’s needs\nand application demands. Data storage, data access, data\n\n\n-----\n\ntransfer, and data visualization activities use distributed computing with low-cost machines to make Big data analysis and\nprocessing possible within a reasonable cost and time [20].\nIn distributed computing systems, the focus is on data representation. It attempts to addresses challenges of processing,\ninteraction, and representation of data in proficient manners.\nInstead of textual/numerical analysis data, visualizations\nare producing better perceptions and understanding of data.\nHowever, power and speed limitations are there when visualizing Big data. This leads to the scalability issue [9]. Current\nprocessing technologies and systems cannot satisfy the needs\nof big growing data, processing of data, and visualization.\nThe increase in speed and storage capacity is much lesser\nthan the amount of Big data. Speed and complex processors\nare needed [24]. Data visualization is a significant way to\ndeal with Big data. It helps to get an overall perspective of\nBig data and find information esteems. It also helps to find the\nhidden patterns in data. The challenge is to manage the parallelism that blends between distributed and shared techniques\nwith such massive data. For the issue mentioned above,\nthe requirements of machines are a fundamental concern.\nMany researchers have shown that large-scale distributed\ndata visualization is an input output-bound problem. When\n\\\ninteractivity is required, security and data access become significant problems, mainly when the data are distributed over\nwide-area networks. Many user interfaces and web-based\ndata visualization techniques are available, which provide\nefficient decentralization with an additional cost for buying servers and equipment. This is the challenge for these\ntechniques. Tables 1 and 2 provide Big data visualization\ntechniques. Table 2 also provides websites to get these tools\nfor the visualization of data. These tools offer visualizations\nof data based on its features. One of the key challenges in\nmodern distributed data visualization today is how to provide\nusers with an interactive experience. Techniques presented in\nTables 1 and 2 are being used in many distributed applications to create and achieve the intended visualization of data\n\n[25].Thesedistributeddatavisualizationframeworksarepresented to attain interactive results and to address scalability.\nDistribution in these techniques is achieved with the help\nof substantial remote servers. Similarly, high-performance\ncomputing capability is needed to get interactive results [14].\nBlockchain is emerging as new distributed technology.\nThe blockchain concept is based on distributed ledger maintained by multiple parties [6]. Using blockchain, we can build\ndecentralized systems [8], which can effectively mitigate\nproblems associated with high access and communication\ncosts. Also, a decentralized blockchain architecture is more\nresistanttoasinglepointoffailure[6].Experimentsshowthat\nthe scalability issue can be effectively handled in blockchain\nsystems.\n\n\n**Table 1 UI-based visualization techniques [25])**\n\nUser interface-based Official websites\n\n1 Tableau Desktop [https://www.tableau.com/](https://www.tableau.com/)\n\n2 Gephi [https://gephi.org/](https://gephi.org/)\n\n3 Jupyter [https://jupyter.org/](https://jupyter.org/)\n\n4 Sisense [https://www.sisense.com/](https://www.sisense.com/ )\n\n5 Qlik [https://www.qlik.com/](https://www.qlik.com/)\n\n6 Team Mate [https://www.wolterskluwer.com](https://www.wolterskluwer.com)\n\n7 Infogram [https://infogram.com/](https://infogram.com/ )\n\n8 Datawrapper [https://www.datawrapper.de/](https://www.datawrapper.de/)\n\n**Table 2 Web-based visualization techniques [25])**\n\nJavaScript based Official website\n\n1 FusionCharts [https://www.fusioncharts.com](https://www.fusioncharts.com)\n\n2 HighCharts [https://www.highcharts.com/](https://www.highcharts.com/)\n\n3 Dygraphs [http://dygraphs.com/](http://dygraphs.com/)\n\n4 Timeline JS [http://dygraphs.com/](http://dygraphs.com/)\n\n5 Chart JS [http://www.chartjs.org/](http://www.chartjs.org/)\n\n6 D3 JS [https://d3js.org/](https://d3js.org/)\n\n7 Leaflet [http://leafletjs.com](http://leafletjs.com)\n\n8 Google Charts www.developers.google.com/chart/\n\n9 RawGraph [http://rawgraphs.io/](http://rawgraphs.io/)\n\nDecentralized computing frameworks are aimed in several ways at disrupting the current cloud environment and\nscalability problem.\nFor achieving interactive outputs from huge streams and\nsets of big data, blockchain-based decentralized approaches\nhave received great attention. Traditional distributed peer-topeer (p2p) networks have inevitable disadvantages, including\ninsecurity and lack of auditing and incentives [12]. Filecoin\n\n[7] and BigchainDB [4] are scalable blockchain databases\nthat combine the characteristics of both blockchain and\nexisting distributed databases. In all respects, blockchain\nis opening doors for solving numerous problems for many\napplications in a distributed manner [2]. Blockchain-based\nframeworks offer many benefits for distributed storage with\nfeatures like availability, no single point of failure, confidentiality, privacy, and integrity [5]. Currently, Filecoin [7], Sia,\nSwarm, and Storj, etc. are mainstream distributed storage\nsolutions built on blockchain [33].\nSia uses blockchain technology to provide an open market to buy and reserve unused computing space for users.\nConditions of storage, such as availability and active duration that are agreed by the participants under file contracts,\nare encrypted service-level arrangements. These are immediately stored on the Sia blockchain and done [29].\nFilecoin is another decentralized blockchain and native\ncloud service. It produces an incentive-based scheme that\n\n## 1 3\n\n\n-----\n\nprovides additional storage. Filecoin uses IPFS, a P2P distributed protocol, where each file is encrypted with a hash\nkey and containing indexing details. It enables large amounts\nof data to be searched and saved and distributed with high\nproductivity [18].\nAnother protocol for decentralized storage network is\nSWARM. It is permission less and communication infrastructure [28]. The principal goal of swarm is to provide the\ndecentralized app’s creators with infrastructure resources. To\nimplement the utilities, SWARM uses smart contract platforms (Ethereum). The strategy for decentralized storage is\na peer-to-peer approach.\nIt is not deniable that blockchain has opened doors for\nsolving many problems for many applications in a distributed\nmanner with provision of various kinds of distributed storage\napproaches mentioned above [2].\nConsidering these blockchains based on distributed storage such as Filecoin, Sia, Swarm, etc, the set of transactions\nare stored within the of blocks of immutable blockchains and\ngenerate a type of decentralized database or storage of structured data. However, due to scalability concerns, the size of\nblocks cannot expand very large, and thus, it can be clearly\nseen that above-mentioned distributed storage blockchains\nare not meant to store and handle large amounts of data and\nwill take too long to process and resources to accomplish\nthese executions. These problems lead to increase the overall energy of the network. Due to consumed energies in nodes\nand long communication distances, the network lifespan, and\nenergy consumption and maximization; it has been a critical\nissue and it increases the overall energy of the network [27].\nHence, these generic schemes are not much energy efficient\nalso do not address optimizing network and storage cost.\nAll projects are similar in the way as they are trying to\nprovide decentralized storage, but existing solutions are controlled by centralized third party and are mainly commercial\napplications. The data stored are available to intermediaries\nwho are providing services. The involvement of intermediaries and third parties leads to an increase in operational and\ntransnational costs and lack of security which makes the system vulnerable.\nHyperledger Sawtooth is also a blockchain framework\nused in developing and running applications in a distributed\nmanner smoothly [13]. Sawtooth also addresses the challenges of permissioned (private) networks. Sawtooth clusters\nwith different approval can be implemented quickly. No central service may leak transaction trends or other classified\ndetails. Sawtooth Blockchain can help to provide efficient\ndatastorageandaccessframeworkformanyprivateandsmall\norganizations and applications [2].\nDespite, all good blockchain systems are facing scalability issues. Due to a number of scalability issues and\nlimited on-chain storage capacity, the feasibility of such a\nblockchain-based data network is restricted [27]. A variety\n\n## 1 3\n\n\nof methods have been suggested to increase scalability while\nmaintaining decentralization and security in this regard.\nSharding technique is one of it.\nSharding is one of the most practical approaches for\nachieving scalability by partitioning network into several\nshards, to reduce the overhead of scalability [32]. However,\nthe issue is to maintain atomicity during cross-shard operations. Each shard is stored in a separate server instance [30].\nAlthough this spreads the load, but increases the overall\ncost and increases the energy consumption. Sharding can be\na good choice when it comes to public and permission-less\nblockchains, but for private small-level projects, sharding\ntechniquemaybecostly,complex,andconsumemoreenergy.\nTable 3 presents a comparison between some blockchainbased storage services. Provided in Table 3 is relative\nestimation four parameters, i.e., closed-source decentralization, exposure of data to third party, network cost and energy\nconsumed, scalability, energy-efficient objectives, and free\nof cost solution.\nMany blockchain-based frameworks offer benefits like\navailability, no single point of failure, confidentiality, privacy, and integrity [5]. Hyperledger Sawtooth is a blockchain\nframework used in developing and running applications in a\ndistributed manner smoothly [13]. Blockchain can provide\nefficient data storage and access framework for many organizations and applications [2]. As the visual representation of\nBig data communicates ideas more clearly, the need for data\nvisualization is increasing in different domains. However, it\nis a complex and demanding task to visualize a massive volume of data. Therefore, a distributed and green solution is\nneeded to achieve interactive data visualization with less cost\nand energy consumption. Using blockchain-based mechanism for distribution and decentralization of big data and then\nvisualize could be a good solution. However, aforementioned\nblockchain-based techniques also increase the chances of\nescalate overall network cost with high energy and resources\nconsumption. Also users may somehow be required to be\ncontrolled by some third-party service making it centralized\nsomehow.\nTherefore, there is need of green solution with minimum\nenergy consumption having low network cost providing optimum storage solution, for small and big levels of projects\nensuring security, available and reliable data, and its visualization at same time.\nBGbV aims to provide support to distributed data visualization. It ensures security and availability, and, most\nimportantly, provides a framework for better utilization of\nexisting distributed resources. By utilizing small units of\nstorage effectively, it facilitates scalability challenges. We\npresent a green solution that will reduce cost by utilizing\nsmall resources that are already available and lessen energy\nconsumption and make it an environmentally friendly solution.\n\n\n-----\n\n**Table 3 Comparison between existing blockchain-based storage services**\n\nParameters Sia Ipfs Swarm StorJ BGbV\n\nClosed-source Yes Yes Yes Yes No\ndecentralization\n\nExposure of data to third Yes Yes Yes Yes No\nparty\n\nNetwork cost and High High High High Low\nenergy consumed\n\nScalability Medium Low–medium Low Low Depends upon\norganization’s free\nstorage resources\n\nEnergy efficiency Not defined Not defined Not Defined Not defined Yes\nobjective\n\nFree of cost adaptation No (SiaCoin) No (FileCoin) Resource trading (No) No (StorJCoin) No\n\n\n### BGbV: blockchain-based green big data visualization\n\nBGbV proposes a distributed architecture of blockchainbased green solution for Big data visualizations. Many\norganizations have loads of distributed storage resources\navailable in unused disk space of workstations and terminals.\nInstead of getting new equipment that requires more hardware and power consumption, this solution provides optimal\nutilization of the existing smaller distributed storage units\nwith the least energy consumption.\nIn BGbV, we have incorporated the hyperledger sawtooth\nframework with our front-end application. Hyperledger sawtooth is highly modular framework and isolates the client part\nof the core system’s application. In our prototype, we have\nconsidered small-to-medium-level organization with midrange resources to store the temporal big data set. According\nto our scope, the permissioned (private) hyperledger sawtooth blockchain is beneficial as it maintain the privacy and\nsecurity. We have developed the client part of our application\nand core system in javascript. Our proposed system presents\nthe design of a blockchain-based data storage and access\nframework for interactive visualization of big temporal data.\nProposed BGbV is different from existing techniques\nmentioned earlier, i.e., due to its integration with blockchain\ntechnology with optimum resource utilization and least\nenergy consumption making it more green solution. It\nensures many significant benefits like effective and optimal\nusage of distributed storage, temper proof record-keeping\nmechanism, no single point of failure, and availability.\nIn BGbV, metadata are stored on the blockchain, whereas\nthe actual files are stored off-chain on multiple locations after\ndistribution. This distribution is according to a pre-defined\nagreement using a peer-to-peer network. This scheme will\nprovide decentralized storage on an economical and secure\nbasis.\n\n\n#### Architecture overview\n\nBGbV framework achieves distributed data visualization\nwith an interactive experience. Distributed processing and\nefficient lookup capabilities are also incorporated in our solution. The major contributions of this research are as follows:\n\n– Design a sawtooth blockchain-based data storage and\naccess framework that stores the metadata on the hyperledger blockchain, whereas the actual data are stored\noff-chain on multiple nodes.\n– Design an off-chain distributed storage data block for\ncost-effective use of blockchain.\n– Distributed and efficient storage of divided blocks of data\non multiple nodes in which contents are accessed through\npermissioned hashes with high throughput.\n– Instead of buying large servers for a remote decentralized solution, employing existing unoccupied storage of\nsmaller machines (nodes).\n– Temper proof record-keeping mechanism.\n– Enforcing quality with no single point of failure and continuous availability of data.\n– An optimized solution toward big data without thirdparty control for decentralization, with minimum network and storage cost.\n\nWehaveintroducedthebasicdistributeddatavisualization\nof temporal data and their interaction with the blockchain. In\nour solution, we have taken a data set for the US’s crime\nrate as Big data in a temporal context. BGbV has two primary stages according to the user’s request and demand\nfor data. It entertains the client’s request based on present\nnodes and node ratings. In BGbV, we have introduced two\nblockchains.BGbV for storing nodes available and their\nresponse-based rating and BGbVfi for storing metadata for\nthe transaction on data.\n\n## 1 3\n\n\n-----\n\n**Fig. 1 Flow of activities in**\n**BGbV blockchain**\n\n**Fig. 2 Flow of activities in**\n**BGbV for information**\nvisualization\n\nIn the first stage, as an initial step, the data set is retrieved\nfrom bulk storage on an administrative view of a web-based\napplication.\nThe flow diagram in Fig. 1 shows the activities perform\non the administrator’s machine to achieve distribution with\nthe help of blockchain. The administrator has the right to\nslice the data on some pre-selected criteria. It can then be\nassigned to an appropriate node that desires to participate\nin storing Big data. Before this assignment of data, a dialogue mechanism is implemented between the administrator\nmachine and nodes. This mechanism ensures that the node\nhas agreed to share its resources and disk space. To store data\nslice, the administrator selects the available node according\nto its storage capacity and rating. It then prompts the storing\nnote by sending a request for sharing space to save data on it.\nAt present, administrator role is defined as an intermediatory\nfor transfer of data from centralized bulk storage to sliced distributed storage. Administrator’s task is to divide the data on\nagreed/pre-defined criteria and send data to storage nodes.\nWhen the data are once stored on agreed nodes, to ensure\nthe integrity, the role of administrator has no role to play in\naccess of data. The role of administrator will be automated\nin future by defining performance metrics and status updates\nof free storage nodes.\nIf this device agrees on saving data on it, the administrator sends the slice of distributed data to that node, and\nits response time is captured. Both the rating and available\ndevices are stored in blockchain BGbV. After this agreement, sliced data are sent to a candidate node. Information\nabout all those nodes that are ready to share space is stored\nin one blockchain BGbV. This blockchain is implemented\nto store two parameters: first is the candidate node identity;\nsecond is its ratings and response time.\n\n## 1 3\n\n\nAfter this process, control is transferred to the second\nblockchain BGbVfi. Multiple copies of same slice are to\nbe distributed among multiple nodes thus providing safeguards against data loss; when data slice is requested the\nnode with better rating and availability is chosen (step 3 of\nFig. 2). The BGbVfi will be used to store the metadata for\ntransactions in our solution. The flow of activities in both\nstages is described with flowcharts in Figs. 1 and 2. We have\nuse used two blockchains for our design prototype instead\nof storing the metadata and ratings in one blockchain. The\nreason for using two blockchains is to maintain the ideal\nscale of blockchain and not to make solution and blockchain\nbulky due to increased number of requests. We have used\ntwo blockchains BGbV and BGbVfi for keeping our transaction metadata log and distribution and slicing information\nseparately. Using two blockchains leads to the benefit of separation of read and write operations from logs and ratings.\nRead/write operation is one operation and metadata of slice\nis other operation. Keeping them separate is a major reason\nfor two blockchains. It disrupts the need for again and again\nreplication of data in block.\nFlowchart in Fig. 1 represents activities initiated in\nresponse to the user’s request for information.\nOur first blockchain BGbVis used to store information\nabout nodes that holds data after agreement mechanism\nand their response-based ratings. Blockchain BGbVfi shown\nin Fig. 2 stores the metadata about the transaction. This\nblockchain will store parameters:\n\n– How administrator sliced data.\n– Amount of data stored on each node with its address.\n– Sending time from administrator’s machine, hash where\nour data have been stored.\n\n\n-----\n\nAbove-mentioned resources are on-chain resources for both\nblockchains. The second stage defines retrieving mechanism.\nFor retrieving mechanism, when the user wanted to retrieve\ndata, the hash and first blockchain are consulted, which helps\nusers check the node with the most ratings and response time.\nThat node is requested to send data to the user. The quality\nof the node is maintained by this blockchain BGbV. The\ninformation that how much data are stored on any node is also\navailableonblockchain **BGbVfi,whichstoredthedistributed**\nmechanism of data.\nThis lets the system choose the amount of data it wants\nto visualize and let them know which node they can get data\nto visualize. When the user sends a request, the node signs\nthe batch and then connects and sends requests to the validator node on the Sawtooth network by using Rest API.\nHyperledger Sawtooth provides a REST API for clients to\ninteract with a validator using common JSON/HTTP standards. REST API uses a JSON envelope to send metadata\nback to clients in an easy and customized manner. The\nsawtooth’s validator sends this transaction request to the\ntransaction processor, which is a blockchain featuring smart\ncontracts. It can apply any rules for the validation of transactions. The transaction processor thus attaches the payload\nto it. The transaction processor makes sure that it is valid, all\ndependencies are met, and then commits it. At present, data\nare sent to the node using FTP which entertains the user’s\nrequest.\n\n#### Off-chain resources\n\nSome off-chain resources will help in the proposed solution. Blockchains are not general-purpose databases. It is not\npossible to store distributed blocks of data as this leads to an\nunmanageable size and causes the problem in the blockchain.\nThus, blockchain only stores metadata, and actual data are\ndivided into blocks and are stored as off-chain resources at\nmultiple storage locations on the network. Therefore, offchain resources contain distributed data on multiple nodes.\nThis adds resilience toward failure and ensures smooth working and high availability. If one of the nodes failed due to any\nreason, availability will be ensured from some other nodes.\nRatings from BGbV are also linked to this decentralized system that gives a score of the node’s reputation. This in term\nhelps the user to better judge if node and their contents should\nbe trusted or not. These off-chain resources can be open to the\npublic or restricted to authorized members. A visualization\npipeline is applied to that data after transfer. The visualization of that data is created for that corresponding user that\nrequests for data visualization. Their computation is thus distributed. Update the state and outcome as a response is sent\nback. Then, the network’s chosen consensus mechanism publishes it in a block to the rest of the network. Figure 3 shows\nthe layered architecture of our given solution.\n\n\n**Fig. 3 Layered architecture of BGbV**\n\n### Analysis of proposed system\n\nIn this section, we present the analysis of the energy and cost\nsaved by BGbV and its adaptation. It is a fact that IT storage\ninfrastructure is under utilized by many organizations. As per\na survey conducted by CISCO [10], the utilization of direct\naccess storage on average is 40% and is 60% for the storage\naccess network. Organizations have underutilized distributed\nstorage available in free disk space on several desktops and\nlaptops. Such organizations can improve the utilization of\nalready available resources by employing BGbV for their\nbig data storage solutions. On average, the cloud-based solution for Big data visualization costs more than 2–4 USD per\nGB per month. Similarly, as Table 3 shows, all decentralized solutions are also commercial solution with associated\ncosts. For most organizations, this cost is besides the cost of\nmaintenance on site underused storage equipment.\nIn the following analysis, we present the details of energy\nconsumed in storing data using BGbV step as compared to\ncloud-based storage. it can be seen that, in terms of energy\nconsumption, the utilization of existing direct access storage\nsolutions consumes less energy than cloud-based storage.\nIn BGbV, we propose two blockchains BGbV and BGbVfi.\nAs mentioned earlier, blockchain BGbV stores the rating\nmechanism, and blockchain BGbVfi saves metadata about\ndata fragments of big data stored at multiple nodes. Each\nblock of BGbV has two parts block header and body. Each\nblock’s body contains transaction data comprising approx.\n264 bytes. The block header’s information contains block\n\n## 1 3\n\n\n-----\n\nnumber, version, previous BlockHash, and the time-stamp\nat which this block is committed. The block’s header is 46\nbytes. Thus, the total size of each block is 310 bytes, which\napproximately equals 0.3 KB. In BGbVfi, the body of the\nblock is 134 bytes. By adding the standard header size of\n46 bytes, the net transaction size is approximately equal to\n0.1 KB. By combining both blockchains of size 0.4 KB to\n0.5 KB, the BGbV takes less than 1 KB overhead for storing\neach fragment.\nFigure 4 shows the amount of energy saved using direct\naccess storage devices compared to the cloud-based solution.\nThe energy is calculated by\n\n_E_ _S = Ecl −_ _E_ _BGbV_ (1)\n\n_Ecl = Fsz ∗_ _λcl_ (2)\n\n� _K_ �\n�\n\n_E_ _BGbV =_ _Fi + K ∗_ _Nnodes ∗_ _Tr_ _BGbV_ ∗ _λH DD. (3)_\n\n_i=1_\n\nHere, E _S is net energy saved, Ecl is net energy consumed by_\ncloud, E _BGbV is net energy consumed by BGbV, Fsz is the_\ntotal number of bits in the file, λcl is the energy consumed\nto write 1 bit on the cloud, λH DD is the energy consumed to\nwrite one-bit disk, Fi are the number of bits in each segment\nstored, Nnodes is the number of nodes maintaining BGbV\nblockchain, K is number to fragments, and Tr _BGbV is the_\nnumber of bits in the transaction. Energy rate λH DD is kept\nas 100FJ and λcl is kept as 1nJ [15].\nFigure 4 shows the energy saved against file size ranging\nfrom 1 to 60 TB is shown. For BGbV, we have sliced each\nfile into fragments of size 1 GB. Also, we have kept the\nnumber of nodes that keep the copy of the blockchain to 100.\nThe red line shows the upward trend in energy saved with\nrespect to file size. Even though the number of fragments\nand transaction overload is also increased (shown as bars),\nbut net energy saved in mJ improves with larger files. The\nabove calculation is performed to establish the bottom line\nof the energy saved by considering just storage. Here, any\nadditional costs of bit transfer to the cloud are not considered.\nWe have calculated only the energy spend in saving original files. It can be seen that backup and replication overhead\nwill show the same trends.\nAlso, for IOT big data, the greatest obstacle for implementing blockchain in IOT is scalability, as the response\nto requests expands as the number of computing machine\nincreases. This leads to high energy consumption. Our solutionwillalsohelpwithIOTbigdata.Asmentiontheproposed\nsystem is decentralized and also use free local underutilized\nstorage will be used to save data. The continuously growing\nIOT data can be efficiently stored and accessed with minimum network energy. BGbV is flexible in terms of source of\ndata. Whether big data are transferred from the cloud or is\ngenerated by IOT, the BGbV framework can easily mange it.\n\n## 1 3\n\n\n#### Features\n\nThe proposed system is intended to ensure many features\nand add support to current distributed data visualization systems. Following are the prominent features of the proposed\nsystem.\n\n**Integrity**\n\nIn the proposed system, the key feature is that all the information is unaltered and immutable, increasing the solution’s\nintegrity. The incorporation of blockchain ensures not only\nintegrity but also provides control over data. Replicas of data\non multiple locations and its metadata as evidence are stored\non the blockchain, ensuring its persistence and integrity.\n\n**Availability**\n\nThe ranking-based node selection in BGbV β provides\nframework for data to be efficiently available for the users.\nUsers can achieve and retrieve that corresponding response,\ni.e., visualization from any other node if one of the part nodes\nis offline. This ensures the data’s availability for visualization\nand analysis from other locations(nodes) on the network.\n\n**Fault tolerance**\n\nBGbV has no single point of failure, and the system is more\ntolerant of faults. The data are vast and divided into different chunks, stored at multiple nodes of the P2P network.\nThis makes our solution more robust and good in terms of\nthroughput.\n\n**Node rating mechanism**\n\nBGbV maintains nodes and their rating with respect to their\nresponsetousers.Eachnodeavailableforstorageisanalyzed,\nand data are collected locally based on the performance. It\ndefines their rating. It also enforces a quality mechanism for\nour proposed system.\n\n**Effective resource utilization**\n\nWhen working with a huge amount of data resource utilization is a very critical concern. It should be logical and\nintuitive. In BGbV, we are not using remote big servers. To\nmake our solution as best economical, we are occupying\nunused smaller units of storage on different machines that\nare available for sharing their space and store data. Instead\nof using high-cost remote servers, data storage is distributed\nand efficient using significantly less cost and smaller units.\n\n\n-----\n\n**Fig. 4 Net energy saved by BGbV**\n\n**Security**\n\nIn BGbV, it is difficult for malicious participants to tamper\nwith the data. Only authorized users have permission and\naccess to data stored, and malicious activities are prevented.\nBlockchain networks also achieve strict security. After passing through a strictly ordered distributed blockchain, transactions are routed to the appropriate transaction processor, and\nhence, this processor then attaches the transactions’ payload.\nBy creating a domain-oriented transaction processor, actions\nare limited that can be performed on a blockchain network,\nwhich can improve security and performance.\n\n### Implementation and testing results\n\nThis section discusses the implementation and the testing\ndetails. This application is developed in HTML 5.0 and Node\nJS as front-end languages.\n\n#### Implementation details\n\nThis section will explain the prototype developed for the\nproof of our concept. Proposed architecture has the following\ncomponents:\n\n– A peer-to-peer network committing transactions between\nnodes.\n– A distributed log, which contains a list of transactions.\n– Atransactionprocessorforprocessingthosetransactions.\n\n\n**Fig. 5 Nodes with details of data stored and response rates**\n\n– An off-chain storage mechanism for storing actual data\nfiles.\n– A distributed storage mechanism for storing the resulting\nstate after processing transaction.\n– A consensus algorithm for ensuring consensus across the\nnetwork.\n\nFor deploying, building, and running blockchain-based green\nBig data visualization (BGbV), we have integrated hyperledger sawtooth framework, which provides a modular and\nflexible platform for implementing transactions between\nnodes. For the proposed BGbV, we have taken m nodes to\nstore the Big dataset of the US crime rate from 2001 to the\npresent. Nodes with details of data stored and response rates\n\n## 1 3\n\n\n-----\n\nare shown in Figs. 4, 5, 6f. The application is developed in\nHtml 5.0 and Node JS as front-end language.\nTheprototypewehaveimplementedishighlymodularand\nhas two basic parts depending on user request and demand\nfor data. The first part is a web-based interface for slicing\nand distributing the data over m nodes. This web interface is\ninitial and mainly runs on the administrator’s machine. The\nadministrator slices the data according to pre-defined criteria\nand sends the data on agreed nodes after a consensus mechanism. The dataset we have taken for this implementation is\nthe US crime rate from 2001 to the present. It is collected\nbased on 20 different attributes of crimes in the US, e.g., to\ncrime types and locations.\nFor our prototype, administrator has sliced the Big dataset\ninto 5 blocks based on these attributes. Each block consists\nof 1000 rows of crime datasets along with its attributes. The\nadministrator can send the sliced data to k nodes agree to\nsave the data and share the space in different replications\nafter consensus.\nWe have created a consensus mechanism in our prototype\nfor running the tests and saving the data on available nodes.\nIt ensures the agreement of nodes and the administrator’s\nmachine to share the space and save data. For this agreement, OAuth 2.0 is used as a protocol for authorization for\nthe specific flow of authentication and authorization for data\nsending to client nodes. We generate a public/private key pair\nfor every available node with their node id for authentication\npurposed to configure the network. For keeping our application highly modular and not bulky, integration of hyperledger\nsawtooth at this stage is kept separately. For incorporating\nSawtooth blockchain, design-specific transaction processor\nTP is written in Node JS. When the node agrees to save data,\nstorage metadata are collected and stored on BGbVfi. At the\nsame time, the response rate is stored on BGbV. The second\npart of our front end application is for the user to retrieve\ndata. While the user wants to retrieve the data, it will select\nnodes with the most higher response ratings from BGbV. A\nREST API is used, which helps users communicate with the\nsawtooth network implemented in our prototype.\nCommunication with the validator and submit any transactions with the help of HTTP requests. A data model\nis designed to tell the allowed operation and transaction\ntypes implemented in the transaction family. For our prototype evaluation, tests are run using the VS code and\ndocker using version 1.49 and version 2.0.0.3, respectively.\nFor our testing scenario, transaction processor runs on\nadministrator machine using a unique machine address “\n**0f5a8cbbf216e93b6b3deed01c7b6866bedb4ed”. For eval-**\nuation and a test run, the administrator has sent the request to\n10 nodes to share their space and hold data. The node receives\nupdates about new blocks from the administrator machine.\nOut of offered, 4 nodes are agreed and ready for sharing space\nand data storage. As measurements for our project, we have\n\n## 1 3\n\n\ncollected the time and response rate of nodes with respect to\ntheir nodes IDs and saved them to BGbV.\nBefore the administrator sends the data, client authentications are established using Oauth2.0 . The administrator\nsends the data to these nodes. Sliced blocks are sent to each\nnode with alternate replication. Node A is saving blocks 1\nand 3, and Similarly, Node B has data blocks 2 and 3. Node\n3 is holding blocks 1,2. Node D, the fourth agreed node,\nis holding blocks 3,2. Once the storage node has received\nthe data, it will verify and send the administrator machine’s\nacknowledgement. After the administrator receives confirmation from the storage node. TP running will commit the\ntransaction are published to BGbV, as shown in Fig. 6f.\nFor testing and retrieving data, user requests node two for\nretrieval of data. The transaction processor also receives the\nrequest sent by the user. Transaction processor assesses the\ntransactiononthisnode.Thetransactionlistinthetransaction\nprocessor is shown in Fig. 6d. Given that all verification pass,\nthe storage node B sends the data back at a high response rate.\nA basic visualization pipeline is associated at this stage, and\ndata visualization is created according to user-requested data\nto visualize data to understand better and get insights into\nthis data. This helps to produce productive output from it.\nDistributed data visualization is presented as DApp using\nhyperledger sawtooth integrated with it. Implementation and\ntesting scenarios are presented, so that the proposed solution\ncan potentially express here.\nFigure 6a, b shows the blocks created; the state list shows\nthat data are saved in blockchain, i.e., metadata. After the\nagreement mechanism and after the verification, the transaction is submitted. Block is added to the blockchain and\nbroadcasted on the network. After this transfer node, grant\naccess to the user requested for data. Transactions committed\nare shown in Fig. 6b.\nFigure 6c shows the successful execution data transfer\nfrom the administrator machine and node. A basic visualization pipeline is associated at this stage, and data visualization\nis created according to criteria the user selected to visualize\ndata to understand better and get insights into these data. This\nhelps to produce productive output from it.\nA distributed data visualization is presented as DApp\nusing hyperledger sawtooth integrated with it. Implementation and testing scenarios are presented to demonstrate the\nproposed solution. Outcome and some of the interactive visualizationsbasedontheuser’srequestareshowninFig. 6e,f.It\nis established from our implementation that data can easily be\nstored in spare storage available in local network, thus reducing overall energy consumptions. The private blockchain\nframework using hyperledger Sawtooth provides support for\ntemper proof record-keeping. Data from centralized storage\nare not only successfully stored on multiple nodes; it is also\nflawless retrieved for visualization.\n\n\n-----\n\n**Fig. 6 Results**\n\n### Conclusions and future work\n\nIn emerging IoT systems, effective resource utilization,\nsecurity, availability, and throughput of Big data are of\nmain concern. We have proposed BGbV, a blockchain-based\nsolution for effective storage and retrieval of Big data visualization. The integration of blockchain solutions provides\ncost-effective, secure, and available visualizations of Big\ndata. BGbV presents a mechanism of the utilization of relatively small distributed resources optimally for storing Big\ndata. As our system provides an alternative to the purchase\nof additional resources, thus reduces energy, costs, and electronic waste. Additional benefits of our scheme are the high\navailability and better extensibility.\nIn future work, the crowd-sourcing of storage nodes will\nbe explored by adding an effective incentive mechanism.\n\n\nNodes will be awarded incentives depending upon the rating and reviews they get from the users. Developing such an\narchitecture that allows large data visualization with incorporating benefits like high availability, no single point of failure,\nquality, and interactive outputs will prove beneficial and will\nnoticeably improve the business sector’s performance.\n\n**Acknowledgements The authors are obliged to the National Uni-**\nversity of Sciences and Technology for funding this work through\nthe Researchers Supporting Grant, National University of Sciences\nand Technology, Islamabad, Pakistan and supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (NRF2021R1A6A1A03039493).\n\n## 1 3\n\n\n-----\n\n#### Declarations\n\n**Conflicts of interest The authors declare no conflicts of interest.**\n\n**Open Access This article is licensed under a Creative Commons**\nAttribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as\nlong as you give appropriate credit to the original author(s) and the\nsource, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material\nin this article are included in the article’s Creative Commons licence,\nunless indicated otherwise in a credit line to the material. If material\nis not included in the article’s Creative Commons licence and your\nintended use is not permitted by statutory regulation or exceeds the\npermitteduse,youwillneedtoobtainpermissiondirectlyfromthecopy[right holder. To view a copy of this licence, visit http://creativecomm](http://creativecommons.org/licenses/by/4.0/)\n[ons.org/licenses/by/4.0/.](http://creativecommons.org/licenses/by/4.0/)\n\n### References\n\n1. Agrawal R, Nyamful C (2016) Challenges of big data storage and\nmanagement. Global J Inf Technol Emerg Technol 6(1):1–10\n2. Ali S, Wang G, White B, Cottrell RL (2018) A blockchainbased decentralized data storage and access framework for pinger.\nIn: 2018 17th IEEE International Conference on Trust, Security and Privacy in Computing and Communications/12th IEEE\nInternational Conference on Big Data Science and Engineering\n(TrustCom/BigDataSE), pp 1303–1308. IEEE\n3. Ali SM, Gupta N, Nayak GK, Lenka RK (2016) Big data visualization: tools and challenges. In: 2016 2nd International conference\non contemporary computing and informatics (IC3I), pp 656–660.\nIEEE\n4. Angeline R, et al. (2018) An immortal database system for the\ndecentralized internet. In: 2018 3rd International conference on\ncommunication and electronics systems (ICCES), pp 994–998.\nIEEE\n5. Asharaf S, Adarsh S (2017) Decentralized computing using\nblockchain technologies and smart contracts: emerging research\nand opportunities. IGI Global\n6. Ayoade G, Karande V, Khan L, Hamlen K (2018) Decentralized\niot data management using blockchain and trusted execution environment. In: 2018 IEEE international conference on information\nreuse and integration (IRI), pp 15–22. IEEE\n7. Benisi NZ, Aminian M, Javadi B (2020) Blockchain-based decentralizedstoragenetworks:asurvey.JNetwComputApplpp102656\n8. Bistarelli S, Pannacci C, Santini F (2019) Capbac in hyperledger\nsawtooth. In: IFIP international conference on distributed applications and interoperable systems, pp 152–169. Springer\n9. Childs H, Geveci B, Schroeder W, Meredith J, Moreland K, Sewell\nC, Kuhlen T, Bethel EW (2013) Research challenges for visualization software. Computer 46(5):34–42\n10. cisco: (2009) [https://www.cisco.com/c/dam/en_us/about/](https://www.cisco.com/c/dam/en_us/about/ciscoitatwork/downloadszciscoitatwork/pdf/Cisco_IT_Ops_Practices_Storage_Utilization.pdf)\n[ciscoitatwork/downloadszciscoitatwork/pdf/Cisco_IT_Ops_](https://www.cisco.com/c/dam/en_us/about/ciscoitatwork/downloadszciscoitatwork/pdf/Cisco_IT_Ops_Practices_Storage_Utilization.pdf)\n[Practices_Storage_Utilization.pdf](https://www.cisco.com/c/dam/en_us/about/ciscoitatwork/downloadszciscoitatwork/pdf/Cisco_IT_Ops_Practices_Storage_Utilization.pdf)\n\n11. Drossis G, Birliraki C, Patsiouras N, Margetis G, Stephanidis C\n(2016) 3d vizualization of large scale data centres. Closer 1:388–\n395\n\n## 1 3\n\n\n12. Huang H, Lin J, Zheng B, Zheng Z, Bian J (2020) When blockchain\nmeets distributed file systems: an overview, challenges, and open\nissues. IEEE Access pp 50574–50586\n[13. Hyperledger: Sawtooth: an introduction (2019). https://www.](https://www.hyperledger.org/learn/white-papers)\n[hyperledger.org/learn/white-papers](https://www.hyperledger.org/learn/white-papers)\n\n14. JavedMU,RehmanM,JavaidN,AldegheishemA,AlrajehN,Tahir\nM (2020) Blockchain-based secure data storage for distributed\nvehicular networks. Appl Sci 10(6):2011\n[15. Jiang W (2018) http://large.stanford.edu/courses/2018/ph240/](http://large.stanford.edu/courses/2018/ph240/jiang2/)\n[jiang2/](http://large.stanford.edu/courses/2018/ph240/jiang2/)\n\n16. KaislerS,ArmourF,EspinosaJA,MoneyW(2013)Bigdata:issues\nand challenges moving forward. In: 2013 46th Hawaii International\nconference on system sciences, pp 995–1004. IEEE\n17. Kosar T (2012) Data intensive distributed computing: challenges\nand solutions for large-scale information management. Inf Sci Ref\n18. Labs P, Benet J (2014) A decentralized storage network for human[ity’s most important information | filecoin . https://filecoin.io/](https://filecoin.io/)\n\n19. Leigh J, Johnson A, Renambot L, Vishwanath V, Peterka T,\nSchwarz N (2012) Visualization of large-scale distributed data. In:\nData intensive distributed computing: challenges and solutions for\nlarge-scale information management, pp. 242–274. IGI Global\n20. Mazumder R, Bhadoria RS, Deka GC (2017) Distributed computing in big data analytics. In: InCon-cepts, technologies and\napplications, Springer, New York\n21. Nguyen BM, Dao TC, Do BL (2020) Towards a blockchain-based\ncertificate authentication system in Vietnam. PeerJ Comput Sci\n6:e266\n22. Olshannikova E, Ometov A, Koucheryavy Y, Olsson T (2016) Visualizing big data. In: Big data technologies and applications, pp\n101–131, Springer, New York\n23. Oussous A, Benjelloun FZ, Lahcen AA, Belfkih S (2018) Big\ndata technologies: a survey. J King Saud Univ Comput Inf Sci\n30(4):431–448\n24. Samal N, Mishra N (2015) Big data processing: big challenges and\nopportunities. J Comput Sci Appl 3(6):177–180\n25. Segall RS, Cook JS (2018) Handbook of research on big data storage and visualization techniques. IGI Global\n26. Seref, Sinanc D (2013) Big data: a review. In: 2013 International\nconference on collaboration technologies and systems (CTS), pp\n42–47, IEEE\n27. Swain A, Salkuti SR, Swain K (2021) An optimized and decentralized energy provision system for smart cities. Energies 14(1451)\n[28. Swarm E (2020) Sawrm documentation . https://swarm-guide.](https://swarm-guide.readthedocs.io/en/latest/introduction.html)\n[readthedocs.io/en/latest/introduction.html](https://swarm-guide.readthedocs.io/en/latest/introduction.html)\n\n[29. Tech S (2020) About-sia . https://sia.tech/about](https://sia.tech/about)\n30. Terado T (2018) What is decentralized storage? (ipfs, filecoin, sia, storj & swarm) . [https://medium.com/bitfwd/](https://medium.com/bitfwd/what-is-decentralised-storage-ipfs -filecoin-sia-storj-swarm-5509e476995f)\n[what-is-decentralised-storage-ipfs-filecoin-sia-storj-swarm-](https://medium.com/bitfwd/what-is-decentralised-storage-ipfs -filecoin-sia-storj-swarm-5509e476995f)\n[5509e476995f](https://medium.com/bitfwd/what-is-decentralised-storage-ipfs -filecoin-sia-storj-swarm-5509e476995f)\n\n31. Yang R, Xu J (2016) Computing at massive scale: Scalability and\ndependability challenges. In: 2016 IEEE symposium on serviceoriented system engineering (SOSE), pp 386–397, IEEE\n32. Yu G, Wang X, Yu K, Ni W, Zhang JA, Liu RP (2020) Survey:\nsharding in blockchains. IEEE Access 8:14155–14181\n33. Zhu Y, Lv C, Zeng Z, Wang J, Pei B (2019) Blockchain-based\ndecentralized storage scheme. J Phys: Conf Ser 1237:042008\n\n**Publisher’s Note Springer Nature remains neutral with regard to juris-**\ndictional claims in published maps and institutional affiliations.\n\n\n-----\n\n"