A Novel Approach for Detecting Malware in Android Applications Using Deep Learning

Detection of android malware depends on the feature vector extraction of android applications statically and dynamically. Static analysis has advantage over dynamic analysis as it covers all source code from byte code and manifest files which contains the permission for applications whereas dynamic analysis of the APK files includes the features like the no. of system calls an application makes, network url it access etc. Feature vector updation in the dataset due to version updates of application creates a challenge for the existing tool to classify the application as malicious or benign. This work creates an combination of neural network and automated tool which collects and updates the feature vectors in the training dataset. This dataset is used by the neural network for its reinforcement training for better classification and classify the application in three classes malicious, benign and can't say. Tensor flow is used for making a neural network which learn from the data extracted in the form of feature vector and classify the applications in the category of malicious, benign or can't say. The work has been able to achieve accuracy of over 80% for a dataset of 1000 sample applications with over 15 different feature vector extracted using designed automated feature vector collection modules.