A hybrid machine learning approach for malicious website detection and accuracy enhancement

Abstract

Malicious URLs are web addresses purposely generated for a user’s detriment. Some examples include phishing scams in which the victim is fooled into logging into a fake site or portals for downloading malware where any click on a link invites a hostile program to the user’s device. The damage done to an individual’s finances, confidential information, and even reputation due to malicious URLs makes it crucial to devise means of countering these threats. This can be achieved by creating an intelligent model that identifies suspicious characteristics common to these websites. The objective of this research is to design a novel hybrid machine learning algorithm-based model for detecting malicious websites. A random forest, decision tree, and extreme gradient boosting (XGBoost) are the three hybrid classification algorithms proposed for the study. Accuracy in detection will help prevent and reduce the effects of such websites. The accuracy rate in this research is 98.7%, precision is at 98.9%, and recall at 98.5%. With these results, it follows that the hybrid model is more effective than training any individual algorithm with the given dataset.