RanDroid: Android malware detection using random machine learning classifiers

The growing polularity of Android based smartphone attracted the distribution of malicious applications developed by attackers which resulted the need for sophisticated malware detection techniques. Several techniques are proposed which use static and/or dynamic features extracted from android application to detect malware. The use of machine learning is adapted in various malware detection techniques to overcome the mannual updation overhead. Machine learning classifiers are widely used to model Android malware patterns based on their static features and dynamic behaviour. To address the problem of malware detection, in this paper we have proposed a machine learning-based malware detection system for Android platform. Our proposed system utilizes the features of collected random samples of goodware and malware apps to train the classifiers. The system extracts requested permissions, vulnerable API calls along with the existence of app's key information such as; dynamic code, reflection code, native code, cryptographic code and database from applications, which was missing in previous proposed solutions and uses them as features in various machine learning classifiers to build classification model. To validate the performance of proposed system, "RanDroid" various experiments have been carriedout, which show that the RanDroid is capable to achieve a high classification accuracy of 97.7 percent.