Malicious URL Detection Using Deep Learning

Abstract

Many businesses rely on the World Wide Web, which is vulnerable to cyber-attacks, newline often starting when users click on malicious links. These deceptive URLs, originating newline from compromised websites, pose threats to the governance, integrity, confidentiality, newline and availability of data. Cybercrime involves using computers to spread malware, illicit content, and other illegal materials. Malicious URLs appear in emails, SMS, social newline media, and pop- ups, tricking users into clicking and exposing them to harmful viruses newline and malware. Existing blocking methods may cause delays, leaving systems vulnerable newline to emerging threats. This thesis introduces two innovative approaches for detecting malicious URLs. The first approach employs Region-Based Support Mapping (RSM) with newline Long Short-Term Memory (LSTM) networks. The process begins with pre-processing, newline which involves tokenization, handling special characters, and word embedding. Significant features are then extracted using a Deep Neural Network (DNN) based on HS newline (Hilbert Space-based). These features are subsequently input into the RSM-LSTM newline model, which classifies URLs as either normal or malicious. The second approach com newlinebines Convolutional Neural Networks (CNN) with LSTM. The pre-processing phase newline includes feature extraction, quantum feature enrichment, and feature scaling. CNN newline identifies key features, which are then passed to the LSTM for classification into normal or malicious categories. Both methods are evaluated based on several metrics, including accuracy, precision, recall, F1 score, ROC, computation time, AUC, true positive rate, newline and false positive rate. The results demonstrate that the proposed models outperform existing ones. The findings contribute to the development of automated threat detection systems and underscore the potential of deep learning in cybersecurity. Additionally, newline the study shows that incorporating specific URL characteristics and behaviors can significantly enhance malicious URL detection, suggesting that the proposed methodology newline