Source

https://www.kaggle.com/datasets/dhoogla/unswnb15?resource=download

  Dataset

This is an academic intrusion detection dataset. All the credit goes to the original authors: dr. Nour Moustafa and dr. Jill Slay. Please cite their original paper and all other appropriate articles listed on the UNSW-NB15 page. The full dataset also offers the pcap, BRO and Argus files along with additional documentation. The modifications to the predesignated train-test sets are… See the full description on the dataset page: https://huggingface.co/datasets/wwydmanski/UNSW-NB15.

This is an academic intrusion detection dataset. All the credit goes to the original authors: dr. Nour Moustafa and dr. Jill Slay.

Please cite their original paper and all other appropriate articles listed on the UNSW-NB15 page.

The full dataset also offers the pcap, BRO and Argus files along with additional documentation.

V1: Original CSVs obtained from here V2: Cleaning -> parquet V3: Reorganize to save storage, only keep original CSVs in V1/V2 V4: Update to remove contaminating features [presentation] & [conference article]

My modifications to the predesignated train-test sets are minimal and designed to decrease disk storage and increase performance & reliability.

In its current iteration, the dataset can be loaded trivially with pd.read_parquet(). All data types are already set correctly and there are 0 records with missing information. Reading parquet files does require fastparquet and / or pyarrow

Exploratory Data Analysis (EDA) through classification with very simple models to .877 AUROC.