Using Machine Learning Algorithms to Detect Malware by Applying Static and Dynamic Analysis Methods

This paper focuses on malware analysis and detection using machine learning methods. The aim of the authors was to perform static and dynamic analysis of programs designed for Windows and then to present the results of the analysis as a dataset. We analysed and implemented different classification methods, such as decision trees, random forests, support vectors and naive Bayes methods. We verified their ability to distinguish malicious and harmless samples and evaluated their success rate using classification accuracy metrics. Then, we compared the results obtained by prediction over the dataset generated by static and dynamic analysis. Classification was more successful on the data gained using the dynamic analysis method. The best malware detection algorithms have been found to be decision tree-based algorithms, in particular the random forest algorithm, which achieves excellent malware detection accuracy of up to 95.95% with a standard deviation of only 0.58%.