Phishing Websites: Subsampled Dataset for Model Training

A 2000-row, 100-column subsample of the PhishingWebsites dataset, generated with a random seed of 0 and stratified sampling. The dataset likely contains features for classifying websites as legitimate or phishing. It was created by Eddie Bergman and is shared under a US public domain license on OpenML.

Use Cases

• Training binary or multi-class classifiers for phishing website detection based on the described features.
• Benchmarking feature selection algorithms on a controlled 100-column subset.
• Studying the impact of stratified sampling on model performance for imbalanced security datasets.

Strengths

• Controlled subsample size of 2000 rows, facilitating reproducible experiments.
• Stratified sampling was applied, which likely preserves the original dataset's class distribution.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is known from the generation parameters, but the original feature meaning and data quality require inspection.
• Last update date is unknown; freshness unverified.

Provenance

Source

OpenML, derived from the 'PhishingWebsites' dataset (ID 4534).

Collection Method

Algorithmic subsampling from a parent dataset using a specified random seed, column limit, and stratified row selection.