Combining Multimodal DNN and SigPid technique for detecting Malicious Android Apps
Balaji Vasu, Neelavathy Pari · 2019
Google announced that they are 2.7 billion active Android users in the world. There are different categories of mobile apps available, based on development nature and different categories, for example apps based on development nature are native apps, hybrid apps, web apps and apps based on different categories are gaming, business, educational, lifestyle, entertainment and utility etc. Due to its scalable growing and open source nature it is easily vulnerable to hackers or attackers to steal our information and inject malware apps into our mobile system. In recent days, a new malware app called Joker was introduced in the mobile market and infected 24 apps in the Google store, which mainly focus on extracting money and credential information from the users without their knowledge. Many scientific researches are undergoing to identify android malicious apps using machine learning classification approach. In this paper, the deep neural network algorithm is chosen to detect Android Malware by extracting features from the Android Manifest file and java API modules then fed into Deep Neural network (DNN) [10] classifier to detect whether the apps as malicious or benign app. This paper combines the Sigpid [1] (Significant Permission Identification) technique which is given as a feature in Multimodal DNN. SigPid was evaluated as a high probability of detecting malware apps using Permissions. The goal is to implement an efficient framework for detecting the Android Malware apps using Machine learning techniques which can be used in App store where all the applications are deployed in cloud and also in Mobile Antivirus app, to detect the harmfulness of the app.