Phishing Detection System Through Hybrid Machine Learning Based on URLs

Abstract

Phishing attacks represent one of the most dangerous forms of cybercrime, exploiting email and websites to deceive individuals and obtain sensitive information. With phishing incidents on the rise, there is a significant need for robust defense mechanisms that leverage advanced machine learning techniques. This study proposes a hybrid LSD (Logistic Regression, Support Vector Machine, and Decision Tree) model aimed at improving phishing detection accuracy and efficiency. The LSD model utilizes both soft and hard voting mechanisms to combine the strengths of the constituent algorithms. For feature selection, we applied the canopy feature selection method, followed by cross-validation and hyperparameter tuning via Grid Search to optimize model performance. The effectiveness of the proposed approach is measured using evaluation metrics such as precision, accuracy, recall, F1-score, and specificity. Comparative analyses demonstrate that the hybrid LSD model significantly outperforms standalone classifiers, including Multinomial Naive Bayes, highlighting its potential in proactive phishing defense.