A Fuzzy Deep Learning Network for Dynamic Mobile Malware Detection

Smartphones and tablets are nowadays targets of malicious writers, that are able to develop more and more aggressive malicious applications to exfiltrate information from mobile devices. The signature-based detection currently exploited in (commercial and free) mobile antimalware is not able to detect never seen threats, as a matter of fact, antimalware are able just to recognise a malware if its signature is stored into the antimalware repository. With this in mind, we propose a mobile malware detector. We consider a dynamic analysis, in particular, we extract system call traces from running applications that, once transformed into images, represent the input for a deep neuro-fuzzy model. The aim of the deep neuro-fuzzy model is to discern malware applications from legitimate ones. We evaluate the deep neuro-fuzzy model effectiveness by considering a dataset composed by 6817 (malware and trusted) real-world Android samples, by reaching a training accuracy of 0.95 and a testing accuracy equal to 0.9, with the aim to empirically demonstrate the effectiveness of the proposed deep neuro-fuzzy model in the Android malware detection task.