ABSTRACT
There has been an exponentially rate of increase in the Android malware samples over the past several years. Reports suggest that such malicious samples pose various severe threats to the user and the smartphone itself such as damage to the system, economical loss to the user, and mobile devices being remotely controlled by servers in the form of mobile botnets. Hence, keeping in mind their threats, several Android malware detection techniques have been proposed in the literature. Many such existing techniques analyze the static attributes for detection including permissions, and intents. Various other techniques exist that aim to use dynamic attributes such as system calls or the network traffic. It has been observed that static techniques are quite easy to implement, however, stealthier samples are successfully able to evade the static detection with the help of update attacks. On the other hand, dynamic solutions were proposed to overcome the limitations of static ones, i.e., to detect stealthy malware. However, again it has been observed by the research community that dynamic solutions increase the computation overhead as compared to static ones. Therefore, in this proposed work, we develop a hybrid detection approach that comprises both static and dynamic attributes for malware detection on Android platform. To achieve this, we rank the attributes according to the Information Gain. Further, we use several machine learning classifiers to test for detection results. The 448experimental results indicate that hybrid model is better than both static and dynamic models.
