Skip to Main Content
Efficient and accurate malware detection is increasingly becoming a necessity for society to operate. Existing malware detection systems have excellent performance in identifying known malware for which signatures are available, but poor performance in anomaly detection for zero day exploits for which signatures have not yet been made available or targeted attacks against a specific entity. The primary goal of this paper is to provide evidence for the potential of learning classifier systems to improve the accuracy of malware detection. A proof of concept is presented for adaptive rule-based malware detection employing learning classifier systems, which combines a rule-based expert system with evolutionary algorithm based reinforcement learning, thus creating a self-training adaptive malware detection system which dynamically evolves detection rules. Experimental results are presented which demonstrate the system's ability to learn effective rules from repeated presentations of a tagged training set and show the degree of generalization achieved on an independent test set.