Skip to Main Content
Since signature based methods cannot identify sophisticated malware quickly and effectively, research is moving toward using samples' runtime behavior. But these methods are often slow and have lower detection rate and are not usually used in antivirus software. In this article we introduce a scalable method that relies on utilizing features other than traditional API calls to obtain higher accuracies. Two feature categories including API names and a combination of API names and their input arguments were extracted to investigate their effect in identifying and distinguishing malware and benign applications. Feature selection techniques are then applied to reduce the number of features and enhance the analysis time. Various classifiers were then utilized along with 10-fold cross validation approach to achieve an accuracy of 98.4% with a false positive rate less than two percent in best case. The small number of extracted features in the proposed technique and the high accuracy achieved makes it an appropriate approach to be used in industrial applications.