On the Efficacy of Static Features to Detect Malicious Applications in Android

The Android OS environment is today increasingly targeted by malwares. Traditional signature based detection algorithms are not able to provide complete protection especially against ad-hoc created malwares. In this paper, we present a feasibility analysis for enhancing the detection accuracy on Android malware for approaches relying on machine learning classifiers and Android applications' static features. Specifically, our study builds on the basis of machine learning classifiers operating over different fusion rules on Android applications' permissions and APIs. We analyse the performance of different configurations in terms of false alarms tradeoff. Results demonstrate that malware detection accuracy could be enhanced in case that detection approaches introduce additional fusion rules e.g., squared average score over the examined features.