Evaluation of Machine Learning Techniques for Malware Detection

Abstract

Currently, there is a security breach in technological systems, attacks on computers and mobile devices through malicious software (also called malware) continue to increase. Malicious software attacks occur at all levels and all types of devices, affecting computer users, corporations, industry, and government. Therefore, the detection of malware continues to be a challenge in computer science. Algorithms based on machine learning and deep learning are being used recently to build software solutions that allow the identification of malicious data in real-time. In this paper, we evaluate three classic machine learning algorithms and one neural network-based algorithm for malware detection and classification using different public data sets. The support vector machine (SVM) and the J48 decision tree (DT) algorithms obtain the best results in all the experiments with values equal to the results available in state-of-the-art. The extreme learning machine (ELM) algorithm obtains very acceptable results in experimentation. The results successfully validate the effectiveness of the implemented algorithms, improving the generalization performance of detecting a new instance of malicious software.