Phishing Website Detection using Supervised ML

With phishing attacks growing in volume and sophistication, automated detection systems need to be both accurate and cost-aware. This project builds a supervised ML pipeline to detect phishing websites using a dataset of 235,794 URLs across 55 features spanning URL structure, domain signals, and page content. After thorough EDA, feature selection was performed using KS Statistic and Mutual Information to isolate the most discriminative predictors. Multiple classification models were trained and compared, with threshold tuning applied to optimize for cost-sensitive outcomes, minimizing the real-world cost of false negatives in a security context. The result is a robust, interpretable detection framework grounded in both statistical rigor and practical tradeoffs.