Charles59/lens-network-traffic

Lens Network Traffic Classification Benchmark

Downstream network-traffic classification data used to evaluate Lens, a knowledge-guided foundation model for network traffic (TMLR). It bundles the 12 classification tasks from the Lens paper as HuggingFace dataset configurations, each with train / validation / test splits and a unified schema.

ℹ️ All tasks are derived from publicly available academic traffic datasets obtained via the NetBench benchmark (Qian et al., 2024); the original terms and citations of those datasets also apply (see Source datasets).

Dataset summary

• Source benchmark: all six underlying datasets are taken from NetBench (Qian et al., 2024).
• Modality: textual renderings of packet/flow sequences (Wireshark/tshark-style lines). Following the paper, each flow is represented by its first packets and classified at flow level.
• Schema (all configs):
  • anonymized_text — string, the (anonymized) packet/flow text used as model input.
  • label — ClassLabel, the class. The stored value is the integer class index; class names are carried in the feature metadata (features['label'].names). One column therefore serves both text-label decoding (few classes) and numeric/digit-label decoding (many classes, e.g. the 196/209-class app-classification tasks — see the paper's note on mapping textual labels to numeric indices to keep the context compact).
• Total examples: ~242,665 across all 12 configs.
• Pretraining data is not released here — only the downstream fine-tuning/evaluation data. (The paper's pretraining split is sampled without any downstream labels to avoid label leakage.)

Tasks / configurations

Task numbers match the Lens paper (Tasks 1–12).

Config (name)	#Classes	Train	Val	Test	Source	Paper task
vpn_detection	2	4,155	1,385	1,385	ISCX-VPN	Task 1
vpn_service_classification	6	13,855	1,728	1,733	ISCX-VPN	Task 2
vpn_application_classification	16	13,859	1,725	1,732	ISCX-VPN	Task 3
tor_service_detection	7	9,033	3,761	3,763	ISCX-Tor	Task 4
ustc-tfc2016_app_detection	16	11,055	10,362	10,366	USTC-TFC-2016	Task 5
crossplatform_android_app_classification	209	6,798	4,196	4,296	Cross Platform (Android)	Task 6
crossplatform_android_app_country_detection	3	6,854	4,283	4,285	Cross Platform (Android)	Task 7
crossplatform_ios_app_classification	196	6,495	2,361	2,456	Cross Platform (iOS)	Task 8
crossplatform_ios_app_country_detection	3	6,531	2,448	2,449	Cross Platform (iOS)	Task 9
dohbrw_query_generator_detection	5	10,909	8,183	8,183	CIC-DoHBrw-2020	Task 10
iot_malicious_detection	2	9,780	8,150	16,301	CIC-IoT-2023	Task 11
iot_method_detection	7	12,605	12,601	12,604	CIC-IoT-2023	Task 12

How to load

from datasets import load_dataset

# pick any config from the table above
ds = load_dataset("Charles59/lens-network-traffic", "vpn_detection")

print(ds)   # DatasetDict({ train, validation, test }) with columns: anonymized_text, label

ex = ds["train"][0]
print(ex["anonymized_text"][:120])
print(ex["label"], "->", ds["train"].features["label"].int2str(ex["label"]))

List all available configs:

from datasets import get_dataset_config_names
get_dataset_config_names("Charles59/lens-network-traffic")

Data fields

Field	Type	Description
anonymized_text	string	Anonymized textual rendering of a packet/flow sequence. Source/destination IP addresses in the flow header are replaced with the special tokens <SIP> / <DIP>. Hex payload bytes and protocol summaries are kept.
label	ClassLabel	Class label. Integer index stored on disk; human-readable class names via features['label'].names / .int2str(i).

Example (vpn_detection)

text : "<SIP> → <DIP> LLMNR 64 Standard query 0xca9a AAAA wpad ca9a 0000 0001 0000 ..."
label: 0  ->  "nonvpn"

Data splits

Splits follow the Lens experimental protocol. A pretrain portion present in some source files has been removed; only train / validation / test are published here. Per-config counts are in the task table. Class distributions are intentionally imbalanced to match real-world conditions (report macro-F1 in addition to accuracy).

Anonymization & privacy

• Flow-level source/destination IP addresses are replaced with <SIP> / <DIP>.
• Anonymization is applied at the flow-header level. Some identifiers embedded inside packet payloads (e.g. IPs in DNS answers / HTTP content, occasional MAC addresses) may remain. The data is released as-is, consistent with the original public datasets it derives from (whose raw payloads are already publicly available).
• No filesystem paths, capture filenames, or raw (non-anonymized) text columns are included.

Source datasets

All datasets are obtained through NetBench (Qian et al., 2024). Please cite the original sources and respect their individual terms of use:

• ISCX-VPN (vpn_*) — Draper-Gil et al., Characterization of Encrypted and VPN Traffic Using Time-Related Features, ICISSP 2016. https://www.unb.ca/cic/datasets/vpn.html
• ISCX-Tor (tor_service_detection) — Habibi Lashkari et al., Characterization of Tor Traffic Using Time-Based Features, ICISSP 2017. https://www.unb.ca/cic/datasets/tor.html
• USTC-TFC-2016 (ustc-tfc2016_app_detection) — Wang et al., Malware Traffic Classification Using Convolutional Neural Network for Representation Learning, ICOIN 2017.
• Cross Platform (Android / iOS) (crossplatform_*) — Van Ede et al., FlowPrint: Semi-Supervised Mobile-App Fingerprinting on Encrypted Network Traffic, NDSS 2020.
• CIC-DoHBrw-2020 (dohbrw_query_generator_detection) — MontazeriShatoori et al., Detection of DoH Tunnels Using Time-Series Classification of Encrypted Traffic, DASC/PiCom/CBDCom/CyberSciTech
  2020. https://www.unb.ca/cic/datasets/dohbrw-2020.html
• CIC-IoT-2023 (iot_*) — Neto et al., CICIoT2023: A Real-Time Dataset and Benchmark for Large-Scale Attacks in IoT Environment, 2023. https://www.unb.ca/cic/datasets/iotdataset-2023.html

License

Released under CC-BY-NC-4.0 (Creative Commons Attribution-NonCommercial 4.0). The benchmark also combines several research datasets, each governed by its own original terms (generally research / non-commercial use with attribution); those terms continue to apply to the respective subsets.

Citation

If you use this benchmark, please cite the Lens paper, the NetBench benchmark, and the relevant source datasets below.

@article{li2026lens,
  title   = {Lens: A Knowledge-Guided Foundation Model for Network Traffic},
  author  = {Li, Xiaochang and Qian, Chen and Wang, Qineng and Kong, Jiangtao and Wang, Yuchen and Yao, Ziyu and Ji, Bo and Cheng, Long and Zhou, Gang and Shao, Huajie},
  journal = {Transactions on Machine Learning Research},
  issn    = {2835-8856},
  year    = {2026},
  url     = {https://openreview.net/forum?id=cGDwTgnJIR},
  note    = {arXiv:2402.03646}
}

@article{qian2024netbench,
  title   = {NetBench: A Large-Scale and Comprehensive Network Traffic Benchmark Dataset for Foundation Models},
  author  = {Qian, Chen and Li, Xiaochang and Wang, Qineng and Zhou, Gang and Shao, Huajie},
  journal = {arXiv preprint arXiv:2403.10319},
  year    = {2024},
  url     = {https://arxiv.org/abs/2403.10319}
}

@inproceedings{drapergil2016iscxvpn,
  title     = {Characterization of Encrypted and VPN Traffic Using Time-Related Features},
  author    = {Draper-Gil, Gerard and Habibi Lashkari, Arash and Mamun, Mohammad Saiful Islam and Ghorbani, Ali A.},
  booktitle = {Proceedings of the 2nd International Conference on Information Systems Security and Privacy (ICISSP)},
  pages     = {407--414},
  year      = {2016}
}

@inproceedings{lashkari2017iscxtor,
  title     = {Characterization of Tor Traffic Using Time Based Features},
  author    = {Habibi Lashkari, Arash and Draper-Gil, Gerard and Mamun, Mohammad Saiful Islam and Ghorbani, Ali A.},
  booktitle = {Proceedings of the 3rd International Conference on Information Systems Security and Privacy (ICISSP)},
  pages     = {253--262},
  year      = {2017}
}

@inproceedings{wang2017ustctfc,
  title     = {Malware Traffic Classification Using Convolutional Neural Network for Representation Learning},
  author    = {Wang, Wei and Zhu, Ming and Zeng, Xuewen and Ye, Xiaozhou and Sheng, Yiqiang},
  booktitle = {2017 International Conference on Information Networking (ICOIN)},
  pages     = {712--717},
  year      = {2017}
}

@inproceedings{vanede2020crossplatform,
  title     = {FlowPrint: Semi-Supervised Mobile-App Fingerprinting on Encrypted Network Traffic},
  author    = {Van Ede, Thijs and Bortolameotti, Riccardo and Continella, Andrea and Ren, Jingjing and Dubois, Daniel J. and Lindorfer, Martina and Choffnes, David and van Steen, Maarten and Peter, Andreas},
  booktitle = {Network and Distributed System Security Symposium (NDSS)},
  year      = {2020}
}

@inproceedings{montazerishatoori2020dohbrw,
  title     = {Detection of DoH Tunnels Using Time-Series Classification of Encrypted Traffic},
  author    = {MontazeriShatoori, Mohammadreza and Davidson, Logan and Kaur, Gurdip and Habibi Lashkari, Arash},
  booktitle = {2020 IEEE Intl Conf on Dependable, Autonomic and Secure Computing (DASC/PiCom/CBDCom/CyberSciTech)},
  pages     = {63--70},
  year      = {2020}
}

@article{neto2023ciciot,
  title   = {CICIoT2023: A Real-Time Dataset and Benchmark for Large-Scale Attacks in IoT Environment},
  author  = {Neto, Euclides Carlos Pinto and Dadkhah, Sajjad and Ferreira, Raphael and Zohourian, Alireza and Lu, Rongxing and Ghorbani, Ali A.},
  year    = {2023}
}