Enhanced Phishing Website Detection Using Dual-Layer CNN and GRU with Attention Mechanism and Lexical NLP Features

This study develops a cutting-edge framework for detecting phishing attacks, leveraging the synergy between natural language processing and advanced deep learning. It begins with a comprehensive dataset and advances through feature engineering to convert textual data into a structured vector space via word embedding, complemented by principal component analysis for efficient dimensionality reduction. This method ensures the preservation of critical analytical information while optimizing computational resources. The framework utilizes NLP to discern linguistic patterns characteristic of phishing attempts, thus enhancing its detection accuracy. The 2CNN-GRU model for phishing website classification effectively combines spatial feature extraction with sequential processing and attention mechanisms to distinguish between benign and malicious websites. A GRU layer is incorporated to process sequential data effectively, capitalizing on its ability to manage information over different time steps, which is pertinent for sequential patterns within URLs. The model's complexity allows it to capture a wide range of features and patterns, improving its accuracy and robustness in phishing detection. K-Fold cross-validation was performed on both the ISCXURL-2016 dataset and the Ebbu2017 dataset. The model exhibited exceptional performance, achieving a 99.91% accuracy rate with the ISCXURL-2016 dataset and a 99.90% accuracy rate with the Ebbu2017 dataset in distinguishing between benign and phishing web entities. These findings highlight the hybrid model's precision in phishing detection, offering significant potential to enhance cybersecurity measures by accurately identifying and mitigating phishing threats.