Research On Intrusion Detection Model Of Fuzzy Min-max Neural Network

With the rapid development of the Internet,cyberspace has become the fifth frontier.More and more cyber attacks have brought huge challenges to cyberspace security.Intrusion detection technology based on various machine learning methods has become an important dynamic security protection method.Fuzzy min-max(FMM)neural network combines fuzzy logic with neural network,and has the advantages of strong adaptive learning ability and real-time performance,but it has the shortcomings of low detection performance,such as high false alarm rate.The particle swarm optimization(PSO)algorithm can optimize the parameters of the model,but it is easy to converge prematurely.Therefore,in this article,the improved PSO algorithm is firstly studied,then it is used to optimize the parameters of the FMM neural network,and finally the intrusion detection model of the FMM neural network is proposed.The main innovations are as follows:1.A particle swarm optimization enhanced with kernel principal component analysis is proposed.The PSO algorithm is easy to converge prematurely.In order to solve this problem,a group search direction is proposed to find the optimal solution from the entire driving swarm instead of considering only the individual and socially optimal particles.The swarm search direction is obtained by a kernel principal component analysis that retains the global nature.Then the speed update strategy of the particles is modified.Experimental results show that the proposed method can avoid premature convergence and enhance the global search capability.2.A redefined fuzzy min-max neural network model is proposed.The classic fuzzy min-max neural network is easy to cause the overlap of hyperboxes from different classes,which affects the classification performance.In order to improve the classification performance,a redefined FMM neural network model is proposed.The basic architecture of the classic FMM neural network is modified.Then,the hyperbox redefinition algorithm is proposed to redefine the min-max points of the hyperboxes.The algorithm includes three processes,namely hyperbox filter,hyperbox optimization and hyperbox combination.Among them,the hyperbox optimization process is realized by the standard particle swarm optimization or particle swarm optimization enhanced with kernel principal component analysis.The redefined FMM learning algorithm is an expansion/contraction/redefinition process.The experimental results show that the classification performance of the model can be effectively improved.3.The intrusion detection model of fuzzy min-max neural network is proposed.The redefined FMM neural network has good classification performance and online learning ability.Therefore,an intrusion detection model based on FMM neural network is proposed.In order to reflect the latest network intrusion behavior patterns and ensure that the detection results are representative,this article uses the UNSW-NB 15 intrusion detection data set with real modern normal behavior and comprehensive attack behavior.The experimental results show that the proposed model has a good detection effect.