AUTOMATED DETECTION OF MALICIOUS URLS USING DEEP LEARNING

Abstract

Phishing attacks have become a serious cybersecurity challenge as attackers continuously create deceptive URLs designed to imitate legitimate websites and obtain sensitive information such as usernames, passwords, financial credentials, and personal data. Conventional phishing detection techniques, including blacklists and manually defined rules, may have difficulty identifying newly generated and previously unseen malicious URLs. To address this challenge, this project presents PhishDetect, a hybrid deep learning approach for intelligent phishing URL classification. The proposed system analyzes the sequential characteristics of URLs and automatically learns patterns associated with phishing and legitimate web addresses. The methodology includes dataset uploading, data cleaning, URL tokenization, numerical encoding, dataset splitting, model training, performance evaluation, and URL prediction. Two bidirectional recurrent neural network architectures, Bidirectional Long Short-Term Memory (Bi-LSTM) and Bidirectional Gated Recurrent Unit (Bi-GRU), are incorporated to capture contextual information from URL sequences in both forward and backward directions. Bi-LSTM is utilized to learn long-term dependencies, while Bi-GRU provides efficient sequential representation with comparatively lower computational complexity. Their complementary learning capabilities are combined to develop a robust hybrid detection framework. The system evaluates classification performance using accuracy, precision, recall, and F1-score and provides a user-friendly web interface for practical testing and prediction. The proposed approach is designed to improve the identification of suspicious URLs while reducing dependence on manually engineered features. Overall, PhishDetect provides an automated and adaptable framework for phishing URL detection that can support safer online interactions and strengthen cybersecurity protection against evolving phishing threats.