BGL-PhishNet: Phishing Website Detection Using Hybrid Model-BERT, GNN, and LightGBM

Abstract

Phishing attacks exploit human and technological vulnerabilities to steal sensitive information, posing a significant threat to online security. To address this challenge, this research introduces a new hybrid method for phishing detection using three advanced techniques: BERT for analyzing texts, Graph Neural Networks for analyzing URLs, and LightGBM for extracting key metadata features. The hybrid approach combines these techniques into a multi-layered model to improve detection accuracy by focusing on text, URL structure, and metadata. The hybrid model achieves an accuracy of 97.3%, outperforming current state-of-the-art models. Experimental results show that this method reduces false positives and enhances phishing detection across various online platforms. The system ensures better real-time performance by integrating outputs from BERT, GNNs, and LightGBM. It also addresses evolving phishing tactics by considering both lexical and structural URL features. This multi-level approach highlights the importance of strengthening online security through robust detection mechanisms.