Construction And Application Of Adaptive Optimization Hyperbox Fuzzy Classifier

Fuzzy minimax neural network(fuzzy hyperbox classifier)is a kind of fuzzy neural network,which has been widely used in medical diagnosis,face recognition or fault classification.Although the fuzzy minimax neural network has strong online learning and classification performance,it still has some shortcomings:(1)The classical fuzzy hyperbox classifier model is constructed in order,and the classification performance is affected by the data input order in the data set.In the classic model of fuzzy hyperbox classifier,the size of the hyperbox is affected by the expansion parameters of the hyperbox,so different data sets need to adjust and set the appropriate expansion parameters.(2)The complexity of the fuzzy minimum and maximum neural network is affected by the number of hyperboxes.The more the number of hyperboxes,the higher the complexity of the neural network,but the classification accuracy of the model will be reduced.Therefore,there is a certain contradiction between the complexity of neural network and the classification accuracy of the model.(3)The super-box expansion operation in the classical fuzzy minimax neural network will lead to the overlap between different types of super-boxes and reduce the classification performance of the model.In view of the limitations of fuzzy hyperbox classifier,this paper proposes the following innovations:1.A fuzzy hyperbox classifier based on two-stage genetic algorithm is proposed.In the fuzzy hyperbox classifier based on two-stage genetic algorithm,genetic algorithm is used to randomly generate multiple extended parameters,and genetic algorithm is used to adjust the extended parameters to find the extended parameter set that makes the model classification performance best.The proposed hyperbox model can reduce the dependence of the hyperbox on the expansion parameters of the hyperbox in the fuzzy hyperbox classifier,and can construct the hyperbox synchronously,and the classification performance of the model is not affected by the input order of the dataset.The experimental results show that the proposed model is superior to the traditional fuzzy minimum maximum neural network model in the accuracy of model classification.2.A fuzzy hyperbox classifier based on fast non-dominated sorting is proposed.In the fuzzy super-box classifier based on fast non-dominated sorting,the multi-objective genetic algorithm of fast non-dominated sorting is used to adjust the relationship between the number of super-boxes in the model and the classification performance of the model,and find the Pareto solution set that can meet both conditions.The proposed model can solve the contradiction between the number of hyperboxes(neural network complexity)and the classification performance of the hyperbox model.The experimental results show that the proposed model is superior to the traditional fuzzy hyperbox classifier model in terms of the number of hyperboxes and the accuracy of the model.3.A fuzzy hyperbox classifier based on differential evolution algorithm and its application in intrusion detection are proposed.In the fuzzy hyperbox classifier based on differential evolution algorithm,the differential evolution algorithm is used to optimize the hyperbox and reduce the overlap between the hyperboxes.Because of the strong classification performance of fuzzy hyperbox classifier,this paper applies the optimized fuzzy hyperbox classifier to intrusion detection.First,the intrusion detection data set is preprocessed,the non-data features are processed numerically,and then the data set is normalized.Finally,the fuzzy hyperbox classifier based on differential evolution algorithm is used for optimization.The experimental results are superior to the classification performance of the previous traditional fuzzy hyperbox classifier model.