PHISH_ATTENTION: achieving robust phishing website detection with balanced datasets and advanced URL features

Abstract The escalating prevalence of phishing attacks in recent years has underscored the imperative need for a comprehensive and sophisticated response to mitigate this pervasive cyber threat. Characterized by deceptive tactics to obtain sensitive information, phishing has evolved in sophistication, resulting in severe consequences such as financial loss, identity theft, and compromise of personal data. This research addresses the inherent challenges faced by existing anti-phishing solutions, encompassing limitations in feature extraction methodologies, suboptimal feature selection, and issues related to dataset imbalance. In response to these challenges, we propose “PHISH_ATTENTION,” an advanced anti-phishing framework that integrates Variational Autoencoders and a Multi-Head Self-Attention Mechanism. The framework is further enhanced by the incorporation of a Deep Convolutional Generative Adversarial Network to effectively address dataset imbalances. Rigorous testing on benchmark datasets reveals that PHISH_ATTENTION achieves a peak detection accuracy of 98.57% with a minimal false alarm rate of 1.09%, surpassing prevailing anti-phishing models. Distinguished by its proficiency in real-time phishing website detection, the framework’s autonomous acquisition of significant URL features establishes it as a resilient and pioneering contribution to the cybersecurity domain.