Malicious URL detection using machine learning techniques

Abstract

With numerous new websites being created every day, it's getting increasingly challenging to tell which ones are safe and which could be dangerous. These websites frequently gather sensitive user data that may be hacked in the absence of proper cybersecurity safeguards, such as the effective identification and categorization of dangerous URLs. In order to improve cybersecurity, this study attempts to create models based on machine learning algorithms for the effective detection and categorization of harmful URLs. In this regard, our proposal uses decision trees, logistic regression, support vector machines, and Naive Bayes to reliably categorize dangerous URLs. To improve classification efficiency, we have integrated hyper-parameter tuning using the Grid Search technique, optimizing model performance for more accurate and reliable results. The results demonstrate the effectiveness of Naive Bayes in achieving high accuracy (91.9%) and reliable performance in detecting malicious URLs. Implementation as a web service of the study provides evidence of the practicality and natural fit into more generalized security frameworks. Ultimately, our approach significantly enhances the detection of unsafe URLs, offering a robust solution to address the growing challenges in cybersecurity.