Smart Phishing Detection on Webpages Using Multi-Agent Deep Learning and Multi-Dimensional Features

Abstract

The increasing sophistication of phishing attacks has made their detection more challenging, as attackers use deceptive tactics to trick users into revealing sensitive information through fraudulent websites. Traditional detection methods struggle to keep up with evolving phishing techniques, necessitating more adaptive approaches. This study introduces a multi-agent deep learning framework that utilizes three specialized models to analyze different aspects of a webpage, including the URL, page content, and Document Object Model structure. The outputs of these models are combined using a highest confidence score mechanism to enhance accuracy. Experimental results demonstrate that this method outperforms existing techniques, achieving 99.21% accuracy with a false positive rate of only 0.22%. It effectively detects both known and new phishing sites, making it a robust solution against emerging threats. Furthermore, this approach highlights the potential of deep reinforcement learning in cybersecurity, paving the way for more automated and resilient security systems to combat phishing attacks.