README.md

metadata

license: cc-by-sa-4.0
task_categories:
  - tabular-classification
language:
  - en
tags:
  - '#BankingData'
  - '#Anti-MoneyLaunderingData'
  - '#AnamolyDetectionData'
size_categories:
  - 1K<n<10K

Banking Data

This is a banking customers, accounts and transaction dataset that can be used for any testing any banking use case such as Anti Money Launder (AML) compliance

Dataset Details

The bank is assumed to be blended bank (good mix of retail and corporate customers) based in the US. The code is parameterized to control the customer/account/transaction characteristics and volume

• Define and prepare master data format, seeded values and the risk levels of the seeded values (manually prepared)
• Define the format for the customer, account and transaction data
• Define exchange rate for the multi-currency transactions
• Define the key thresholds and conditions
  • number of customers, accounts, transactions needed
  • Risk thresholds: To control the high/medium/low risk composition of customers and transactions
  • Allowed combinations for ‘product-transaction type’, ‘channel-transaction type’
  • Transaction type for Funds Transfer International
  • Currency Threshold: To control the composition of currency of the transactions

Master data used – Business of the customer (Business Code), Geography, Customer Type, Product, Transaction Channel, Transaction Type and Currency. These master data is prepared manually and each of the values for these master data is mapped to indicate the risk level. For HR-CP scenario all these master data is used except Transaction Type and Customer Type which are useful if the scope is expanded to cover other AML scenarios

Transaction data – Customer, Account and Transaction data are prepared synthetically using python scripts and Faker library. Effort is taken to ensure the transaction data reflects the production data to the extent possible with (a) volume thresholds based on currency, risk level (b) Right ratio of customers to accounts to transactions (c) Dependency between product-transaction type, channel-transaction types, currency-transaction types

A total of 400 customers, 1000 accounts and 10000 transactions are generated synthetically with a transaction date spanning around 15 days.

I have used this in an AI model developed for AML compliance and the details are in the Models section <>

Files

• BankingData-SyntheticData.ipynb – This file Colab notebook containing the code for generating customer, account and transaction data as mentioned above
• Input Files
  • MasterData.xlsx, Conditions.xlsx - Two master data files which are input to the synthetic data generation
• Output Files
  • customers.csv - has customer data for 400 customers
  • accounts.csv - has customer data for 1000 accounts
  • transactions.csv – has transaction details for 10000 transactions over 14 transaction days
  • transactiionFeatures.csv - transaction file enriched with more features like counterparty business code, PEP flag etc which are needed for further analysis / modelling of transaction data