Robust Malware Detection Using Machine Learning

Malware is a growing global internet threat. The anti-malware systems are the primary defense mechanism against malwares. In this paper we propose a robust malware detection method which detects malware without having any textual information about the URLs of websites. This makes identifying potential malware sites for domain registrars easier, efficient and faster without considering complex sets of features such as website logos and URLs. We use several machine learning algorithms and compare it to previous research. The results indicate that Gradient has the lowest error rate of 11% compared to other models, while Naive Bayes should be used when our specific purpose is to detect malware websites only.