Phishing Detection in Dynamic Environments Using Network Behavior

Phishing is a growing cybersecurity threat, prompting the need for effective detection systems. This study introduces a deep learning framework that combines CNN, LSTM, and Transformer layers to identify phishing websites by analyzing URL features such as URL and domain lengths, the inclusion of an IP in the domain, TLD, HTTPS presence, and character ratios. The CNN model reached an accuracy of 93%, the LSTM model obtained 85%, and the Transformer model surpassed both with an accuracy of 99%, a recall of 100%, and an F1 score of 99%. The proposed layered model achieved 97% accuracy, proving its effectiveness in detecting phishing websites. By incorporating CNN, LSTM, and Transformer layers, the model captures sequential relationships and emphasizes important data trends, providing a powerful tool for phishing detection and enhancing online security.