| license: mit | |
| language: | |
| - en | |
| - ru | |
| tags: | |
| - http-requests | |
| - waf | |
| - cybersecurity | |
| - parquet | |
| # MicroData | |
| A reproducible byte-oriented HTTP request dataset for defensive WAF research. It contains **800,000** records with 60% benign and 40% attack examples, split into train (70%), validation (15%), and test (15%). | |
| ## Fields | |
| - `raw_bytes_b64`: UTF-8 request bytes encoded as base64. | |
| - `label`: `0` benign, `1` attack. | |
| - `family`: attack family or `benign`. | |
| - `source`: source provenance. | |
| - `obfuscation_level`: `none`, `light`, `medium`, or `heavy`. | |
| ## Sources and construction | |
| The mix includes CSIC 2010-derived HTTP traffic, PayloadsAllTheThings payload references, the MIT-licensed `notesbymuneeb/ai-waf-dataset`, and deterministic synthetic examples. Mutations include URL/double encoding, case changes, comment insertion, null/whitespace tricks, Unicode-safe textual variants, and injection-location changes. Seed: `20260921`. | |
| ## Loading | |
| ```python | |
| from datasets import load_dataset | |
| ds = load_dataset("neuroX3/MicroData") | |
| ``` | |
| This dataset is intended for defensive model training and evaluation. Review source licenses and provenance before redistribution in downstream products. | |