PhishHook: Catching Phishing Schemes using Machine Learning

Abstract

Phishing attacks remain a formidable threat in today's digital landscape, posing significant risks to individuals and organizations. The ever-evolving nature of these attacks outpaces conventional detection methods, demanding innovative solutions. This paper introduces a cutting-edge dual-layer model for phishing authentication, leveraging the combined power of Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks on raw URL data. Our approach begins with the meticulous gathering and cleansing of diverse, up-to-date URL datasets to ensure a comprehensive foundation for analysis. The self-designed CNN extracts spatial patterns inherent in phishing URLs, while the LSTM network captures temporal dependencies and contextual nuances, significantly enhancing detection accuracy. This hybrid model achieves an impressive 98% accuracy, outstripping traditional machine-learning techniques in precision and recall. Extensive experimentation confirms the superiority of our model, which not only minimizes false positives and negatives but maintains computational efficiency, making it suitable for real-time deployment. The study underscores the critical need for continuous dataset updates and model retraining to keep pace with emerging threats, ensuring robust protection in an increasingly perilous cyberspace. This work represents a significant advance in phishing detection, offering a scalable, high-performance solution that meets the challenges of today's dynamic threat environment.