Abstract: Phishing continues to be one of the most deceptive and persistent cyber threats in today’s interconnected digital landscape, targeting unsuspecting users through fraudulent websites and emails designed to steal sensitive information. This study presents a robust hybrid deep learning model that integrates Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) architectures to enhance phishing detection accuracy and resilience. The CNN layers effectively capture spatial and lexical patterns from URLs and email content, while the BiLSTM layers analyze sequential dependencies and contextual relationships within textual data. Together, these components enable the model to learn both structural and semantic cues associated with phishing behavior.Experimental evaluations conducted on benchmark phishing datasets demonstrated that the proposed hybrid CNN-BiLSTM model achieved an overall detection accuracy exceeding 95%, outperforming traditional machine learning algorithms such as SVM and Random Forest. The system also showed superior precision and recall, reducing false positives and improving interpretability through an integrated attention mechanism. This research contributes to the advancement of cybersecurity by proposing an adaptive, data-driven defense framework capable of evolving alongside emerging phishing strategies and offering practical potential for real-time threat mitigation.
Keywords: Phishing Detection, CNN, BiLSTM, Deep Learning, Cybersecurity.

Downloads: | DOI: 10.17148/IJIREEICE.2025.131042