Image-Based Malware Detection Through a Deep Neuro-Fuzzy Model

From ransomware, able to cipher data and ask for a ransom to have back access to own files, to spyware, developed to collect private and sensitive information regarding the user's activity without his consent, in the current malware landscape there is a huge variety of malicious payloads aimed to perpetrate damage. Current antimalware, mainly based on the signature principle, require that the signature of a threat must be stored in the antimalware database, in order to identify (and try to remove) the infection: for this reason, they are unable to detect zero-day threats. Considering that in the literature there are several methods adopting machine learning to detect malware, in this paper we propose a deep neuro-fuzzy network for malware detection. An accuracy equal to 0.935 is obtained in the evaluation of more than 20000 real-world malware (belonging to 10 different categories) and trusted applications, demonstrating the effectiveness of the proposed deep neuro-fuzzy model for malware detection.