<![CDATA[Phishing attacks conducted through fraudulent URLs continue to become a significant cybersecurity challenge, demanding detection methods that are both rapid and reliable. This research introduces a hybrid deep learning model that integrates Convolutional Neural Network (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) architectures for automated phishing URL detection. The CNN component is utilized to identify lexical and character-level patterns within URLs, whereas BiLSTM captures contextual dependencies from both forward and backward sequence directions. The proposed model was trained using 450,176 labeled URL samples with character-level tokenization, minimizing the need for manual feature extraction. Experimental evaluations demonstrated excellent performance, achieving an accuracy of 99.78%, while precision and recall consistently remained above 99% according to confusion matrix analysis. Furthermore, the trained model was implemented into a Python-based check_link() function that integrates deep learning prediction scores with rule-based analysis to categorize URLs into safe, suspicious, or dangerous classes. Nevertheless, inference testing revealed that the model is sensitive to incomplete URL formats, particularly legitimate domains lacking the www prefix. In general, the CNN-BiLSTM approach proved highly effective for large-scale phishing URL classification, although incorporating additional external features may help reduce false positive predictions.Keywords: Phising; URL; Deep learning; Convolutional Neural Network; Bidirectional Long Short-Term Memory.]]>
Copyrights © 2026
