DaE2: A Diverse and Efficient Ensemble Framework for Real-Time Phishing URL Detection

Abstract

Phishing attacks continue to pose a significant cybersecurity threat, with over 1.2 million unique phishing sites detected in 2023. Traditional detection methods struggle to keep pace with rapidly evolving phishing techniques, particularly given the short lifespan of modern phishing websites. This paper presents DaE2 (Diverse and Efficient Ensemble), a novel ensemble-based approach for real-time phishing URL detection. Our framework integrates four distinct ensemble methods—AdaBoost, Bagging, Stacking, and Voting—each chosen for their unique strengths in handling different aspects of URL classification. Using a comprehensive dataset of 11,055 URLs with 30 engineered features, we demonstrate that DaE2 achieves 98.7% accuracy while maintaining real-time processing capabilities. The framework shows particular strength in detecting newly created phishing URLs (99.1% accuracy) and handling sophisticated masquerading techniques (98.5% accuracy), while maintaining low false positive rates (1.6%). Performance analysis reveals that our approach outperforms traditional methods by 2.5% in accuracy while reducing processing time by 35% compared to deep learning approaches. The framework's adaptive weighting mechanisms and parallel processing implementation ensure robust performance across varying types of phishing attacks, making it suitable for real-world deployment.