Malicious URL: Analysis and Detection using Machine Learning

Malicious URLs (Uniform Resource Locator) are known for hosting unsolicited content that can deceive common users and lead to scams resulting in significant financial losses, private information theft, and malware installation. Despite the development of prevention strategies, it remains a significant risk, even with measures such as receptive blacklisting of URLs. Unfortunately, this approach falls short due to the short lifespan of phishing websites. To address this problem, a real-time malicious URL detection method is required, which can effectively detect and prevent such attacks. Typically, phishing URLs have certain connections between the registered domain level and the URL's path. The approach presented in this paper uses these connections to characterize the URLs by inter-relatedness, and various attributes were mined from them to make estimations. These attributes were then used in several machine learning techniques to detect malicious URLs from a real dataset. Maximum accuracy of 90% was achieved using K-N earest Neighbors Algorithm. The patterns for a URL being Malicious or Benign were also visualized using graphs.