Android Applications Classification with Deep Neural Networks

Abstract

In an effort to foster the growth of Android in the mobile operating system market and keep current consumers, Google has made millions of applications, some of which are free and others of which are paid apps, available in the Google Play store. Users have, however, regu- larly complained that the store is full of malicious apps and low quality apps, putting their devices and personal information at risk. Detec- tion of mobile applications vulnerabilities remains a significant challenge due to the constant evolution of methods to obfusticate and circumvent current detection and security schemes.The ability to correctly classify and categorize mobile applications, especially those built for Android, is crucial for separating malignant applications from benign ones thereby protecting the many more devices of unsuspecting users.This paper presents a deep neural network technique to classify android applications into legitimate and malware applications. Specifically,we first proposed applications classification model based on deep belief neural network classifier.The neural network was built and trained on real dataset to classify android-based applications using TensorFlow library and imple- mented on python programming language. We further trained and tested our neural network’s classification performance against that of four tra- ditional deep feed-forward neural networks and seven baseline models based on machine learning algorithms on the same data. According to experimental results, a deep belief neural network-based model could accurately categorize Android apps into benign and malicious cate- gories with 98.7% of the time. Compared to all previous deep learning and machine learning methods, this represents a significant improve- ment. Also,the categorization accuracy of the DBN model is better than that of numerous other models examined by earlier researchers.