Lightweight malware detection based on machine learning algorithms and the android manifest file
Monica Kumaran, Wenjia Li · 2016
This study aims to learn if the Android manifest file provides enough information to classify an app as malicious or benign. In particular it compares the efficacy of using requested permissions versus inter-app intent communication. It also improves static malware detection by comparing and refining different machine learning algorithms on the manifest file dataset. I find that a Cubic Support Vector Machine (SVM) algorithm is the most accurate classifier for the complete dataset with a 91.7% accuracy. I also find that combining intent filters with requested permissions improves the classifier, but intent filters are not enough to base a classifier on while permissions are.