A Hybrid Malware Detection Model For Android Operating System Using Deep Learning

The Android operating system has grown rapidly in the past ten years and has become the world's largest operating system since its inception.The outstanding achievements of its rapid occupation of the smart mobile device market inevitably attracted great attention from malware makers and researchers in related fields.The Android operating system allows users to download and install software from various channels rather than the only official application market.This provides great convenience to malware makers.So they can spread various forms of malware targeting user's smart mobile device through thirdparty application stores.The number of malware targeting the Android operating system has exploded in the past few years.Despite the continuous advancement of malware detection technology,there are numerous cases caused by various types of malware,which has greatly affected the privacy and property security of users.Therefore,accurate and effective detection of software on the Android operating system,filtering out malicious software,and then maintaining the software quality of the Android application market and user information security have become very urgent and of great significance.Most existing malware detection technologies use static analysis,dynamic analysis,and machine learning algorithms for malware detection and classification.Among them,static and dynamic analysis methods have a high false positive rate.The accuracy of traditional machine learning algorithms depends heavily on manual selection of massive features by researchers which cannot cope with the current trend of big data.In response to the above problems,this thesis proposes a multi-layered Android malware detection system based on multi-modal deep neural networks.In this thesis,feature information is extracted from application samples by static analysis.Then a multi-modal network structure was used to receive inputs of feature vectors.So that all data samples and multiple features extracted from it can be effectively used.At the same time,the multi-modal input method greatly improves the scalability of the system in the feature input section.By using deep neural networks,the complex multi-layered structure is used to gradually extract and mine deep features from the mass behavior features one by one,so as to provide highly relevant feature data for malware detection.The experimental results prove that the deep neural network-based Android malware multi-layer detection system can effectively and accurately detect malware.Compared with other models,it can make more sufficient use of samples and multiple features to improve the overall detection capability.The system update speed is much faster than other models.So it can deal with the current situation of the massive increase of Android applications more effectively.Also,the experiment proves that the system has detection ability and response ability to a certain extent against zero-day malware.