PhishGuard ML: Evaluating Machine Learning Models for Phishing Message Detection

PhishGuard ML is an independent machine learning research project evaluating text-based classification methods for identifying suspicious email messages. Using more than 5,600 messages from the Apache SpamAssassin public mail corpus, the study compares TF-IDF–based Logistic Regression and Linear SVM models, with particular attention to false negatives, model errors, and cross-source generalization. The study evaluates performance using accuracy, precision, recall, F1 score, ROC-AUC, and PR-AUC, and includes feature and error analyses to examine patterns associated with successful and missed classifications. A source-held-out experiment is also used to assess how well the models transfer to separately released email data. Because the SpamAssassin corpus provides spam rather than independently verified phishing labels, the positive class is treated as a spam/suspicious-email proxy rather than a phishing-specific ground truth. The project is intended as an empirical study of machine learning classification behavior and generalization, not as a production-ready phishing detection system. Code and experimental materials are available in the accompanying GitHub repository. https://github.com/SujayO372/PhishGuard-ML