An Ensemble Deep Learning Based Detection System to Identify Phishing Attacks

Abstract

Phishing websites pose a significant threat in the cyber security realm, resulting in substantial financial losses. Despite ongoing updates to confrontation methods, the effectiveness of these approaches remains unsatisfactory. The proliferation of phishing websites in recent years has raised concerns, highlighting the urgent need for more advanced phishing detection technology. Machine learning and deep learning support defensive mechanism to protect from phishing attacks. A recent study introduced DeepPhish, a deep neural network to generate phishing URLs, demonstrating the evolving sophistication of phishing tactics. The expansion of phishing beyond traditional channels like email, SMS, and pop-ups to include mobile platforms and social networks has made detection more challenging. Phishing attacks encompass QR code phishing, spear phishing, and spoofing on mobile applications. Furthermore, malicious actors often host phishing sites on HTTPS and SSL-certified domains to predict them as legitimate. This diversification of phishing tactics presents new obstacles detection. Despite the elusive nature of phishes, security experts and researchers made significant efforts to combat phishing website threats. To address this issue, we propose an ensemble deep learning-based detection system to identifying both AI-generated and human-crafted phishing URLs. By incorporating URL HTML Encoding for enhanced lexical analysis, our system can classify URLs in real-time and compare them with existing detection methods. In this proposed model enhanced phishing web page detection model by CNN and Bi-Directional LSTM to identify phishing attacks.