Ensemble Feature Selection and Machine Learning Approaches for Phishing Website Detection Using

Phishing, a prevalent social engineering attack, targets users' sensitive information by mimicking legitimate websites to deceive users into divulging personal data. Despite various detection techniques, zero-day phishing attacks remain challenging to identify. This study aims to improve phishing detection through advanced feature selection methods and ensemble machine learning techniques. We propose a novel approach that integrates multiple feature extraction models, including LIME, Kernel SHAP, Tree SHAP, and Information Gain, to enhance feature selection and improve classification accuracy. Three datasets —ISCX-URL-۲۰۱۶, Hannousse & Yahiouche, and UCI —were utilized for evaluation. The ensemble learning techniques of voting and stacking were employed, combining classifiers including Random Forest, XGBoost, and Support Vector Machine (SVM). Results indicate that our stacking-based model outperforms traditional methods, achieving high accuracy across various datasets. This approach demonstrates the potential for robust detection of phishing websites with reduced computational complexity.

کلیدواژه ها:

Phishing ، Machine Learning ، Ensemble ، Feature Selection