Phishing Websites Subset for Model Training with 2,000 Rows and 100 Features

2,000 rows of data derived from the PhishingWebsites dataset on OpenML, subsampled using a random seed of 4. The subset likely contains 100 features and up to 10 target classes, generated by Eddie Bergman. The original dataset is in the public domain (us-pd).

Use Cases

• Training binary or multi-class classifiers for phishing website detection based on URL and page features.
• Benchmarking feature selection algorithms on a dataset with 100 potential attributes.
• Studying the impact of stratified sampling on model performance for imbalanced security datasets.

Strengths

• Subsampling parameters are explicitly documented: seed=4, nrows=2000, ncols=100, nclasses=10.
• The generation code indicates stratified sampling was used, which can help preserve class distribution.
• Derived from a known source dataset (PhishingWebsites ID 4534) on a reputable platform (OpenML).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown for the original source, and the subsample size may limit statistical power.
• Last update date is unknown; freshness unverified.

Provenance

Source

OpenML dataset 'PhishingWebsites' (ID 4534), subsampled by Eddie Bergman.

Collection Method

Algorithmic subsampling with random column and row selection, potentially using stratified sampling.