Comparative analysis of machine learning algorithms for phishing detection

Phishing attacks have become one of the most prevalent forms of cybercrime, leading to significant financial losses and breaches of personal information. Traditional rule-based methods of detecting phishing websites and emails are increasingly insufficient due to the evolving sophistication of attackers. Machine learning (ML) provides a promising alternative by enabling automated classification of phishing and legitimate instances based on extracted features.
This study presents a comparative analysis of five widely used ML algorithms, namely Decision Tree, Random Forest, Support Vector Machine (SVM), Naïve Bayes, and Logistic Regression, for phishing detection. A publicly available phishing dataset was utilized, containing both legitimate and malicious samples with relevant URL and website-based features. Preprocessing steps included feature encoding and normalization. The models were evaluated using standard performance metrics: accuracy, precision, recall, F1, score, and ROC, AUC. The results indicate that ensemble-based models, particularly Random Forest, achieved superior performance across most metrics, with higher accuracy and robustness against overfitting compared to single classifiers. While Logistic Regression and Naïve Bayes offered lightweight alternatives with faster training times, their predictive power was comparatively lower. The findings highlight the importance of algorithm selection in phishing detection systems and provide practical insights for cybersecurity practitioners. Future work will extend this analysis by incorporating larger datasets and exploring deep learning approaches for real-time phishing detection.