Advanced Phishing Website Detection with SMOTETomek-XGB: Addressing Class Imbalance for Optimal Results

Abstract

In the evolving field of cybersecurity, detecting phishing websites poses a unique challenge, primarily due to the issue of class imbalance. This research proposes a novel approach by combining the SMOTETomek resampling technique with the XGBoost classifier to tackle this challenge. SMOTETomek, a hybrid technique that combines SMOTE with Tomek Links, addresses both oversampling and undersampling by enhancing minority class representation while eliminating ambiguous instances. The proposed SMOTETomek-XGB model consistently surpasses traditional classifiers across key performance metrics, including accuracy, F1 score, recall, precision, and ROC-AUC. This combination significantly improves phishing detection, advancing the state of the art in mitigating online threats. The results suggest that SMOTETomek-XGB is an essential tool for enhancing detection capabilities in cybersecurity.