Improving Malware Detection with a Novel Dataset Based on API Calls

Abstract

In this paper, we analyze current methods to distinguish malware from benign software using Machine Learning (ML) and feature engineering techniques that have been implemented in recent years. Moreover, we build a new dataset based on API calls gathered from software analysis, conforming more than 30000 samples belonging to malware as well as benign software. Finally, we test this dataset with an existing model that achieves accuracy rates close to 97% with a different, smaller dataset, identifying interesting results that can open new research opportunities in this field.