Android Malware Detection Using Local Binary Pattern And Principal Component Analysis

Since Android has become the most popular smartphone operating system in the world,the number of Android malicious applications are increasing rapidly as well.These malicious applications may make users' privacy information such as account details,contact information,photographs,etc.easy to leakage without their knowledge,which bring incalculable spiritual and economical losses to the users.Due to the endless evolution of Android malware,it is critical to make some research on Android malware detection.In the past few years,some techniques based on machine learning were proposed to prevent,thwart and detect Android malware.Basically they were combined with static analysis or dynamic analysis.However,both static analysis and dynamic analysis based on machine learning have some shortages.For example,static analysis needs to decompile apps to get its features and dynamic analysis needs to monitor the behavior of app and trace what an app really does as well,which are all time consuming.Also,static analysis may occupy local resources while decompiling and dynamic analysis may also occupy network resource when uploading applications on the cloud.Moreover,there do exist some scenarios without any decompiler,sandbox or virtual machines,etc.,which cannot detect Android malware any more.In view of the shortcomings of static analysis and dynamic analysis,this paper propose a novel Android malware detection method based on Local Binary Pattern(LBP)and Principal Components Analysis(PCA),we firstly read Android binary directly and convert to a gray image representation,which is used for LBP texture feature extraction to get feature vector,and then,we use PCA to reduce the dimension of Android binary texture feature vectors,in the last,we classify malware and benign by K Nearest Neighbor(KNN).The main contributions of our work are summarized as follows.Initially,We propose a novel approach for Android malware detection based on binary texture feature,which can visualize Android application direactly and extract texture feature for Android malware detection.Secondly,our approach can efficiently detect Android malware without any decompiler,sandbox or virtual machines.Thirdly,experimentation on 5560 malwares and 5127 benigns shows that our approach can achieve high detection accuracy and greatly reduce detection time cost and resource occupation at the same time.