puyang2025/waf_data_v2

WAF Data v2

Dataset Summary

This dataset contains HTTP request records with labels for security-related classification tasks. Each row includes request metadata and content fields, plus a string label.

Supported Tasks and Leaderboards

• text-classification: classify HTTP request records based on the label field

Languages

• English (en) for textual tokens in headers/body; protocol and method fields use standard HTTP tokens

Dataset Structure

Data Instances

Each instance is a single HTTP request:

• method: HTTP method (string)
• url: request URL (string)
• protocol: HTTP protocol/version (string)
• headers: serialized headers (string)
• body: request body (string)
• label: classification label (string)
  • normal: benign/legitimate request
  • anomalous: suspicious or attack-like request

Data Files

The dataset is stored as Parquet files with the following splits:

• train.parquet: 1,924,461 rows
• eval.parquet: 240,560 rows
• test.parquet: 240,558 rows
• heldout.parquet: 147,188 rows

Data Fields

All columns are stored as strings in the Parquet schema:

• method (string)
• url (string)
• protocol (string)
• headers (string)
• body (string)
• label (string)

Data Preparation

JSON source files were converted to Parquet for efficient storage and access. The Parquet files are the canonical dataset artifacts.

Licensing Information

MIT License.

How to Load

from datasets import load_dataset

data_files = {
    "train": "train.parquet",
    "validation": "eval.parquet",
    "test": "test.parquet",
    "heldout": "heldout.parquet",
}

ds = load_dataset("parquet", data_files=data_files)

Citation Information

If you use this dataset, please add your citation here.