Comparison of Machine Learning Models for Identifying Malicious URLs
Manasi Dnyanesh Karajgar, Sonali Sawardekar, Seema Khamankar, Nalini Tiwari, Manjusha Indrajit Patil, Vishal Borate, Yogesh Kisan Mali, Ashvini Chaudhari
Abstract
Malicious URLs, posing threats such as phishing and malware, cause significant financial losses globally. Detecting these URLs automatically before users access them is essential for cyber security. This study explores various machine learning techniques to accurately identify malicious URLs. Decision trees, random forests, K-nearest neighbors (KNN), and naive Bayes models are assessed using a dataset containing over 700,000 URLs. Ensemble models like random forest and extra trees exhibit superior performance, achieving over 93% accuracy in distinguishing between benign and malicious URLs. However, class imbalance remains a challenge, with minority malicious types often exhibiting lower precision. Comparative analysis highlights the feasibility of employing ensemble machine learning for automated malicious URL detection. With adequate examples and feature engineering, tree-based models can effectively identify threats and bolster cyber defense.