A comparative deep learning approach for phishing URL detection

Phishing remains a major and serious challenge in this digital era. Recently, deep learning- based approaches have demonstrated superior performance in accurately identifying and classifying phishing URLs. This paper examines and compares modern deep learning methods, with a focus on RNNs and their variants. Moreover, Cybersecurity faces a noteworthy challenge from phishing attacks. Most existing studies rely either on content-based methods, which generalize poorly to new domains and often struggle with previously unseen URLs, or on conventional machine learning methods that require substantial domain expertise and intensive manual feature design. In contrast, deep learning techniques offer a promising alternative for effective phishing URL classification. The Gated Recurrent Unit (GRU) model demonstrated superior performance in phishing URL classification tasks. Its ability to capture sequential patterns effectively makes it a valuable tool for combating phishing attacks in online environments Experimental evaluation conducted on a dataset of 60,000 URLs shows the Gated Recurrent Unit (GRU) model consistently surpasses the standard RNN and other recurrent variants including LSTM, BiLSTM, and BiGRU; in terms of accuracy, precision, recall, AUC, as well as both training and testing efficiency. Further research in this area has strong potential to enhance the effectiveness and efficiency of cybersecurity defenses against phishing attacks. Consequently, the GRU model emerges as a preferred option for future research on phishing URL classification.