Research On Android Malware Detection Technology Based On Deep Learning

Android is a Linux-based operating system in intelligent mobile device that is completely open and has high market share.In recent years,the number of Android malware has increased rapidly because of the complex and diverse Android application market and the openness of the Android system.due to the widespread use of obfuscation technology in malware,traditional detection methods have a poor detection effect on obfuscated malware.This paper proposes an Android malware detection model based on deep learning.By automatically extracting features of Android application and using deep learning models to detect Android malware,the accuracy of Android malware detection is improved.The main work of this article is as follows:(1)Android application feature set is constructed with obfuscation-invariant ability.This paper finds that traditional static feature set don't have obfuscation-invariant ability after studying the security framework and traditional malware detection technology of the Android system and obfuscation technology used by Android malware.On the basis of traditional static feature set,this paper adds new static features with obfuscation-invariant ability,and extracts the dynamic features during application's runtime.As a supplement to the static feature set,the dynamic features enrich the Android application feature set.(2)A deep learning model based on deep belief networks and gate recurrent unit is proposed.This paper analyzes the feasibility of using deep learning algorithms for Android malware.Because of the difference of static features and dynamic features,this paper uses two different deep learning models: deep belief network and gate recurrent unit separately.The output is connected to the fully connected layer at the same time and the classifier fine-tunes the network in a supervised manner to achieve model convergence.(3)The effectiveness of the Android malware detection model based on deep learning is tested by experiment.The experimental results show that deep learning model based on deep belief network and gate recurrent unit achieves 96.75% detection accuracy of the obfuscated malware,which is superior to traditional machine learning algorithms and deep belief network model.It proves that the Android malware detection model based on deep learning proposed in this paper has a good detection effect.