Enhanced Phishing URL Detection Using a Novel GRU-CNN Hybrid Approach

As cybercriminals become their tactics, phishing URLs are increasingly operated to exploit unsuspecting users. This leads to notable financial loss and erodes user trust in online systems, influencing businesses and individuals. Though effective in specific scenarios, traditional signature-based and heuristic methods often require help keeping pace with the dynamic of phishing schemes. In this study, we introduce a hybrid model combining Gated Recurrent Unit (GRU) and Convolutional Neural Networks (CNN) to enhance phishing URL detection. Our primary purpose was to utilize both temporal feature extraction through GRU and spatial feature extraction using CNN, building a robust model capable of effectively identifying phishing attempts. We evaluated three models, GRU, CNN, and the proposed GRU+CNN hybrid, employing a Kaggle dataset containing over 2.5 million URL samples labeled as phishing. The GRU model reached 97.8% accuracy, while the CNN model performed slightly better, with 98% accuracy. However, the hybrid GRU+CNN model outperformed, achieving an accuracy of 99.0%, showing its superiority in addressing the complexities of phishing detection. Future work will optimize the hybrid model for real-time detection and investigate its adaptability to other cybersecurity domains, such as malware and social engineering threats.