Detecting malicious programs is becoming increasingly more difficult due to the variety of behaviors and forms malware nowadays presents.
Hash-based and heuristic driven approaches used by current antivirus seem not to address efficiently nor effectively the problem. To address the problem Machine Learning based approaches are increasingly more used with notable success. In this paper we want to analyze such approaches to assess the accuracy of this ML based methods, in particular in a supervised learning setting, providing the reader a systematization of knowledge on the state of art in the field.