Enhanced Cyber Phishing Website Detection using a Stacked Classifier Combining ML and DL

Phishing, a prevalent cyberattack, deceives individuals into disclosing sensitive details or installing malware. Initially executed via email, phishing attacks have expanded to text messages, phone calls, and social media. Despite cybersecurity advancements, phishing remains a major threat, with 3.4 billion phishing emails sent daily. In response, we propose a hybrid stacked classifier combining ML and DL models for phishing website detection. The first layer includes Random Forest and Decision Tree, while the second layer integrates Multi-Layer Perceptron and XGBoost. The model achieves a test accuracy of 97.38%, precision of 97.14%, recall of 98.10%, and an F1 score of 97.62%. Performance is further validated by ROC analysis, demonstrating its effectiveness in minimizing false positives and maximizing true positive rates. This approach offers a robust, scalable solution for phishing detection, with potential for real-time application optimization.