Polymorphic Malware Detection based on Supervised Machine Learning

Abstract

Currently, the size of malicious software grows faster each year and poses a thoughtful global security threat. The number of malware developed is increasing as computers became interconnected, at an alarming rate in the 1990s. This scenario caused a rising number of malware. It also caused many protections are developed to against the malware. Malware authors have created them to be more challenging to be evaded from anti- virus scanner. Extracting the behaviour of polymorphic malware is one of the major issues that affect the detection result. The main objectives in this work are to extract the best feature selection and to increase detection of polymorphic malware. This study used machine learning to improve malware detection accuracy. This research used four types of machine algorithm which are K-Nearest Neighbours, Decision Tree, Logistic Regression, and Random Forest. As with most studies , careful attention was paid to false positive and false negative rates which reduced their overall detection accuracy and effectiveness. The result showed that the Random Forest algorithm is the best detection accuracy compares to others classifier with 99 % on a relatively small dataset. The implementation of a feature selection technique plays an important role in machine learning algorithms to increase the performance of polymorphic malware detection