Android platform takes 85% of the global OS market share, where it runs multiple types of devices, smart phones, wearables and all other different shapes. As a result, android be- comes one of the top cyber criminals targets to spread their malwares. There are traditional approaches used to detect malwares through code and behavior analysis, but recently new alter- native approaches have been introduced. In this project, we use one of the recent approaches, i.e. Image-based analysis and the machine learning techniques. Also, since android applications (APK) consists of multiple file, this project aims to provide a comparative study on using a single or more of these files in order to determine which file has an impact in characterizing a malware. The process of detecting a malicious APK starts with visualizing one or more of the APK’s files into gray-scale image, then extracting the image features using GIST descriptor, and finally training using machine learning Random Forrest classifier to detect the malicious APK.