Research And Implementation Of Multi-Classifier Joint Android Malware Detection Technology Based On Weight Assignment

Since the beginning of the 21 st century,major breakthroughs have been made one after another in research fields such as real-time communication.Massive application software based on mobile terminal devices has facilitated daily life.However,there are a large number of malicious application software hidden in the application software market,which brings unnecessary losses to users.At present,people’s concern has changed from what new functions the application software achieves to whether the application software can be used "safely".The change of concerns means that the solution to the application software security problem is imminent.The Android operating system,as the most used intelligent system in the world,is the first to bear the brunt.By using Android malware detection technology,malicious application software can be effectively avoided from flowing into the application software market.However,in the face of the explosive growth of the number of application software,manual analysis has been difficult to deal with.In the past decade,more and more research work has applied machine learning technology to detect malware.Although the application of machine learning technology has greatly improved the efficiency of detection,the detection effect of the classifier is also limited by the machine learning algorithm used for modeling.With the change of detection data and application scenarios,the detection performance of the classifier will also be change.Therefore,while improving the detection accuracy,the detection method should also have better generalization ability.Based on the above viewpoints,this paper studies the working mode of the collaborative detection of multiple base classifiers,designs a decision mechanism that matches the working mode to assist in the detection to achieve the goal of multi-based classifier collaboration,and realizes the multi-based classifier collaborative detection framework.The contributions of this paper are as follows:1)On the basis of investigating the existing Android malware detection methods,aiming at the problem that a single machine learning algorithm shows different performance in different application scenarios,a weighted voting detection method based on the collaboration of multiple machine learning base classifiers is designed,and The detection process of the whole method is described.Based on the characteristics of different algorithm base classifiers suitable for different application scenarios,the method further integrates the judgment results of multiple base classifiers with the decision mechanism of weighted voting,so that the method can adapt to more complex application scenarios and improve the detection efficiency.Accuracy.The experimental verification shows that the method can effectively detect the properties of Android application software,and each performance index is better than the base classifier working independently,and it also achieves better results compared with other detection methods.2)The existing researches on Android malware detection based on deep learning are investigated.Aiming at how to combine deep learning with Android malware detection technology more effectively,a multi-deep learning classifier collaborative detection method based on maximum joint value is designed according to the characteristics of deep learning classification mode,and the detailed detection steps are given.The method avoids the overfitting problem in deep learning by modeling base classifiers of different architectures,and combines the maximum joint value determination mechanism to statistically analyze the determination results of the base classifiers,which improves the detection performance and ensures the detection efficiency.The designed experiments show that the method can identify the nature of application software with a high accuracy rate.Compared with a single deep learning classifier,all indicators have been improved,and compared with some existing detection methods,it also has certain advantages.3)The architecture design and functional realization of the weighted voting detection method based on the cooperative work of multiple machine learning classifiers and the multi-deep learning classifier cooperative detection method based on the maximum joint value are carried out,and the implemented example systems are tested.The final test results show that the system can detect Android malware in different modes according to the different needs of users,and can get a good detection effect.