AI-Driven Phishing Website Detection: An Empirical Evaluation of Machine Learning Algorithms and Feature-Based Classification

Abstract

Phishing websites are a major cybersecurity threat because they imitate legitimate websites to deceive users into revealing sensitive information, including passwords, financial details, and personal data. Traditional phishing detection techniques, such as blocklists and manually defined rules, may fail to identify newly created or rapidly modified malicious websites. Artificial intelligence (AI) and machine learning (ML) offer a data-driven approach to identifying suspicious websites through the analysis of URL characteristics and other website-related features. This paper presents an empirical evaluation framework for comparing machine learning algorithms for phishing website detection. The proposed study uses the publicly available UCI Phishing Websites dataset, which contains 11,055 instances and 30 predictive features. Four supervised learning algorithms—Logistic Regression, Decision Tree, Random Forest, and Support Vector Machine—are selected for comparative evaluation. The methodology incorporates data preprocessing, stratified training and testing, cross-validation, feature-importance analysis, and multiple performance metrics, including precision, recall, F1-score, and ROC-AUC. Existing research indicates that ensemble learning methods, particularly Random Forest, are promising for this classification task; however, performance depends on dataset characteristics and experimental design. Since no new model execution was conducted for this manuscript, numerical experimental results are not claimed. The proposed framework supports reproducible evaluation and highlights the importance of temporal validation, interpretable predictions, and real-world testing in developing reliable phishing detection systems.