Android malware detection based on multi-category and tag level ratio

Abstract

The arrival of the mobile platforms has brought much convenience to our life, people start to get used to social events, online shopping, and business activities with mobile phones, especially those with the Android system. Since the open source of Android system, it has rapidly penetrated into various internet ecosystems, a large number of Android applications are released to the market every day. However, more and more malicious applications are being developed to steal user privacy and property. In this paper, firstly, based on the AndroidManifest file structure, a tag level ratio feature is proposed for detecting Android malicious applications. Secondly, a new hybrid feature selection strategy is used to preprocessing permission and API features. Finally, through single model and ensemble model, 1100 malicious applications and 1567 nonmalicious applications were detected. The result show that our method has a good accuracy.