A Machine-Learning-Based Framework for Supporting Malware Detection and Analysis

Malware is one of the most significant threats in today's computing world since the number of websites distributing malware is increasing at a rapid rate. The relevance of features of unpacked malicious and benign executables like mnemonics, instruction opcodes, API to identify a feature that classifies the executables is investigated in this paper. By applying Analysis of Variance and Minimum Redundancy Maximum Relevance to a sizeable feature space, prominent features are extracted. By creating feature vectors using individual and combined features (mnemonic), we conducted the experiments. By means of experiments we observe that Multimodal framework achieves better accuracy than the Unimodal one.