An adaptive neuro-fuzzy inference system for multinomial malware classification

Malware detection and classification are important requirements for information security because malware poses a great threat to computer users. As the growth of technology increases, malware is getting more sophisticated and thereby more difficult to detect. Machine learning techniques have been extensively used for malware detection and classification. However, most of them are binomial classifications that only detect the presence of malware but do not classify them into types. This study sets out to develop a multinomial malware classifier using an adaptive neuro-fuzzy inference system (ANFIS) and investigate the effectiveness of ANFIS in the classification. A first-order Sugeno ANFIS model was developed. It has five layers and uses two if-then rules. The ANFIS model was trained and tested with two prominent malware datasets from the Canada Institute of Cyber Security. The experimental results showed that the performance of the ANFIS model degrades as the size of the datasets increases, and the accuracy, precision, recall, and root mean square error is 94%, 0.88, 0.87, and 0.19 respectively.