Experimental Evaluation of Description Logic Concept Learning Algorithms for Static Malware Detection

Abstract

In this paper, we propose a novel approach for malware detection by using description logics learning algorithms. Over the last years, there has been a huge growth in the number of detected malware, leading to over a million unique samples observed per day. Although traditional machine learning approaches seem to be ideal for the malware detection task, we see very few of them deployed in real world solutions. Our proofof- concept solution performs learning task from semantic input data and provides fully explainable results together with a higher robustness against adversarial attacks. Experimental results show that our solution is suitable for malware detection and we can achieve higher detection rates with additional improvements, such as enhancing the ontology with a larger amount of expert knowledge.