Machine learning based malware classification for Android applications using multimodal image representations

The popularity of smartphones usage especially Android mobile platform has increased to 80% of share in global smartphone operating systems market, as a result of which it is on the top in the attacker's target list. The fact of having more private data and low security assurance letting the attacker to write several malware programs in different ways for smartphone, and the possibility of obfuscating the malware detection applications through different coding techniques is giving more energy to attacker. Several approaches have been proposed to detect malwares through code analysis which are now severely facing the problem of code obfuscation and high computation requirement. We propose a machine learning based method to detect android malware by analyzing the visual representation of binary formatted apk file into Grayscale, RGB, CMYK and HSL. GIST feature from malware and benign image dataset were extracted and used to train machine learning algorithms. Initial experimental results are encouraging and computationally effective. Among machine learning algorithms Random Forest have achieved highest accuracy of 91% for grayscale image, which can be further improved by tuning the various parameters.