An Efficient Framework for Phishing URL Detection Using Adaptive Drift‐Aware Feature Monitoring and Online Learning
Sruthi Krishnan, S. Manohar Naik · Security and Privacy · 2025
ABSTRACT The relentless evolution of phishing methodologies causes significant financial losses, data breaches, and reputational damages for individuals and organizations. Traditional detection methods struggle with limitations such as the inability to adapt to rapidly changing attack patterns and the computational inefficiency of frequent model updates. To address these limitations, we propose a novel framework with the following objectives: develop an adaptive framework integrating dynamic feature selection with online learning techniques, implement efficient drift detection mechanisms for real‐time feature distribution monitoring, and validate performance improvements against conventional methods. Our proposed method employs Kolmogorov‐Smirnov tests for drift detection alongside online Stochastic Gradient Descent for continuous model optimization. Validation using real‐time phishing URLs from PhishTank demonstrates 99.85% accuracy with a minimal 0.067% false positive rate, substantially outperforming traditional approaches.