README.md

---
license: mit
language:
- en
tags:
- intrusion-detection
- network-logs
- cybersecurity
pretty_name: Network Traffic Dataset
---

# Network Traffic Detection Dataset

A curated dataset for **network traffic anomaly detection**, derived from the **[BCCC-CSE-CIC-IDS2018](https://www.yorku.ca/research/bccc/ucs-technical/cybersecurity-datasets-cds/large-scale-intrusion-detection-dataset-bccc-cse-cic-ids2018/)** intrusion detection dataset, which itself is an enhanced version of [CSE-CIC-IDS2018](https://www.unb.ca/cic/datasets/ids-2018.html)  
This dataset is restructured for modern machine learning and deep learning research on **network security** and **intrusion detection**.

> [!IMPORTANT]  
> The raw dataset (~90GB) was fragmented across 34 `.csv` files. To optimize for ML workflows, the data was merged and compressed into `.parquet` chunks using the `polars` Python library.  
> During the merge, 18 columns exhibited data-type mismatches (e.g., containing both `float64` values and strings like *"not a complete handshake"*). To resolve this, these specific columns were type-cast to `Utf8`.  
> - [View Schema Comparison Output](https://huggingface.co/datasets/init5iv3/network-traffic-detection/raw/main/schema_comparison_output.txt)  
> - [View Merge Script](https://huggingface.co/datasets/init5iv3/network-traffic-detection/raw/main/merge-upload.py)

## Usage

> [!IMPORTANT]
> This dataset is large (~90GB uncompressed). \
> Attempting to load the entire dataset into memory at once (e.g., via `pd.read_parquet()` or `dataset.to_pandas()`) will cause Out-of-Memory (OOM) crashes on some machines. \
> To safely use this dataset, separate the download step from the ingestion step, and process the data using lazy evaluation or batching.

### Accessing & Downloading the Data
Choose the method that best fits your system's constraints: \
_Hugging Face docs on [downloading datasets](https://huggingface.co/docs/hub/datasets-downloading)_

[datasets](https://huggingface.co/docs/hub/datasets-usage) \
_downloads the dataset using Apache Arrow_
```python
from datasets import load_dataset

dataset = load_dataset("init5iv3/network-traffic-detection", split="train")
```

[streaming with datasets](https://huggingface.co/docs/datasets/v4.8.4/en/stream) \
_Iterate through the data over the network without downloading the entire dataset to disk_
```python
from datasets import load_dataset

dataset = load_dataset("init5iv3/network-traffic-detection", split="train", streaming=True)
```

local download via [CLI](https://huggingface.co/docs/hub/datasets-downloading#using-the-hugging-face-client-library) \
_download `.parquet` files to a local directory_
```bash
hf download --quiet init5iv3/network-traffic-detection --repo-type dataset --local-dir /path/to/dir
```

local download via [snapshot_download](https://huggingface.co/docs/huggingface_hub/en/guides/download#download-an-entire-repository) \
_download `.parquet` files to a local directory_
```python
from huggingface_hub import snapshot_download

snapshot_download(
  repo_id="init5iv3/network-traffic-detection",
  repo_type="dataset",
  local_dir="/path/to/dir",
  allow_patterns="*.parquet",
  resume_download=True
)
```

virtual mounting with [hf-mount](https://github.com/huggingface/hf-mount)
```bash
hf-mount start repo init5iv3/network-traffic-detection /path/to/local/mount
```

### Ingesting/Processing

[polars](https://huggingface.co/docs/datasets/use_with_polars) \
_if the dataset was downloaded locally, `polars` can lazily scan the fragmented `.parquet` files in chunks_
```python
import polars as pl

df_lazy = pl.scan_parquet("/path/to/dir/**/*.parquet")

label_counts = df_lazy.select("label").value_counts().collect()
```

batch-processing with `datasets` library
```python
from datasets import load_dataset

dataset = load_dataset("init5iv3/network-traffic-detection", split="train")

for batch in dataset.iter(batch_size=10000):
  # process features here.
  pass
```


## Overview

Raw intrusion detection datasets like BCCC-CSE-CIC-IDS2018 are large, fragmented across multiple files, and often difficult to use directly for ML experiments. \
This dataset was created to provide an ML-ready format to the raw data, containing over 300 network flow features extracted via [NTLFlowLyzer](https://github.com/ahlashkari/NTLFlowLyzer). \
It is suitable for binary and multi-class classification.

## Labels

### Binary Classification
1. **Benign**
2. **Non-benign:** Aggregated from all specific attack classes below 

### Multi‑Class Classification
- Benign  
- DoS-Hulk
- DoS-Slowhttptest
- DoS-GoldenEye
- DoS-Slowloris
- DDoS-LOIC
- DDoS-HOIC
- Brute-Force-XSS
- Brute-Force-Web
- Brute-Force-FTP
- Brute-Force-SSH
- SQL-Injection
- Botnet
- Infiltration

## Citation

If you use this dataset, please cite both this repository and the original source:

### This dataset

```bibtex
@misc{init5iv32026networktraffic,
  title={Network Traffic Dataset},
  author={init5iv3},
  year={2026},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/init5iv3/network-traffic-detection}
}
```

### Original dataset (BCCC-CSE-CIC-IDS2018) and paper

```bibtex
@misc{bccc_cse_cic_ids2018_dataset,
  title={BCCC-CSE-CIC-IDS2018: Large-Scale Intrusion Detection Dataset},
  author={Shafi, MohammadMoein and Lashkari, Arash Habibi and Roudsari, Arousha Haghighian},
  year={2025},
  publisher={Behaviour-Centric Cybersecurity Center - York University},
  url={https://www.yorku.ca/research/bccc/ucs-technical/cybersecurity-datasets-cds/large-scale-intrusion-detection-dataset-bccc-cse-cic-ids2018/}
}

@article{shafi2025toward,
  title={Toward Generating a Large Scale Intrusion Detection Dataset and Intruders Behavioral Profiling Using Network and Transportation Layers Traffic Flow Analyzer (NTLFlowLyzer)},
  author={Shafi, MohammadMoein and Lashkari, Arash Habibi and Roudsari, Arousha Haghighian},
  journal={Journal of Network and Systems Management},
  volume={33},
  number={44},
  year={2025},
  publisher={Springer}
}
```

## License

This dataset is released under the **MIT License**.  
The original **BCCC-CSE-CIC-IDS2018** dataset is subject to its own licensing terms.

## Contributions

Feedback is welcome via the [community page](https://huggingface.co/datasets/init5iv3/network-traffic-detection/discussions)