URL under Siege: A Comparative Analysis of Malicious URL Identification with Advanced Techniques

Safeguarding the confidentiality and integrity of users' private information is a significant challenge in cyber security, particularly due to the rapidly evolving landscape of technology and threats. Traditional malicious Uniform Resource Locator (URL) detection methods, such as blacklist-based approaches, often fail to identify emerging threats. In contrast, Machine Learning (ML) and Deep Learning (DL) algorithms leverage large datasets to analyze URL features and behaviors, enabling more accurate classification of URLs as malicious or benign. Recent research demonstrates that deep learning models, such as Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNNs), these models eliminate the need for extensive manual feature engineering by automatically extracting relevant features from raw URL data. Furthermore, hybrid approaches combining multiple algorithms show promise for enhancing detection performance. This study emphasizes the importance of developing ensemble techniques. The ensemble CNN +DNN technique accuracy, precision, recall and F1 score is 88.00%, 85.00%, 82.00% and 83.00% respectively.