A Comprehensive Review Of Phishing Detection Techniques Based On Machine Learning

Abstract

Phishing attacks continue to pose significant security threats to individuals and organizations worldwide, resulting in financial losses and compromised sensitive information. This comprehensive review examines various machine learning (ML) techniques employed for detecting phishing attempts across multiple vectors, including websites, URLs, and emails. By analysing recent literature, we explore feature selection methodologies, prominent algorithms, dataset characteristics, and performance metrics. Our findings indicate that supervised machine learning approaches, particularly Random Forest and Convolutional Neural Networks, demonstrate superior detection accuracy, often exceeding 97%. Traditional ML algorithms combined with effective feature selection techniques provide practical solutions with reasonable computational requirements, while deep learning approaches offer higher accuracy at the cost of increased complexity. Notable research gaps include limited attention to zero-day attacks, insufficient multimodal phishing detection techniques, and ethical considerations surrounding privacy and consent. This review provides valuable insights for security researchers and practitioners seeking to advance the state-of-the-art in phishing detection through machine learning.