Research On Virus Detection Technique Based On Ensemble Neural Network

With the rapid development of science and technology,the computer has become an indispensable tool,but the computer virus has caused great inconvenience to people.Therefore,achieving a common virus detection model has great practical significance.This article firstly analysis the related principle of computer virus and neural network,and elaborates on the characteristics,composition,classification of computer virus and the features of neural network,and discusses the theory of neural network ensemble in detail,and then uses improves BP classifier and D-S evidence theory as a new integration technology to optimize neural network integration.In order to select higher recognition rate and larger difference of classifiers to improve the generalization ability of the integration system,this article uses two ways to improve BP classifier.One way is to introduce the momentum factor which technology is mature,and the other one is to propose improved adaptive learning rate on the base of traditional adaptive learning rate.In this paper,the improved BP classifier not only can ease the convergence speed and the minimum value problem,but also can effectively increase the difference degree between the classifiers.For voting unfairness of ensemble neural network,introduce D-S evidence theory to fuse multiple sources as result,and improve the ratio of recognition without increasing the algorithm complexity.This paper presents a virus detection model based on improved ensemble neural network and model can effectively identify a variety of common virus and it is a generic virus detector.Sample feature extraction uses N-gram model for the first screening and uses information gain for the secondary screening,in order to reduce the amount of input information and information dimension.Training process uses Bagging algorithm to disturbance training data,in order to improve the difference degree between child classifiers.Combining the hardware and software development environment of the laboratory,this paper design a simulation experiment on the Matlab platform.Experiments prove that the improved BP algorithm is improved on the problem which is convergence speed and minimum problems.Experiments show that the detection effect of the integrated detection effect is better than single classifier.With the traditional model of integrated neural network virus detection based on voting,experiments verify the detection effect of this model that proves the accuracy and reduces the rate of false positive.