Research On SVM Feature Selection And Parameter Optimization Based On Improved Fireworks Algorithm

With the rapid development of the Internet,computer network brings great convenience to people with its features of spreading information quickly,open and freely.At the same time,users have to face the privacy protection when sharing resources.Especially,in recent years,the promotion of applications such as e-commerce and e-government has made cybersecurity becomes a major problem for large corporations and research institutes.When it comes to network security,the first thing that people think of is the firewall,access control and other technologies.However,these technologies have some limitations.The main disadvantage is that they cannot defend themselves from the internal network attacking.In fact,almost half of all serious attacks and intrusion come from internal users.Intrusion detection technology to make up for this shortcoming was put forward.It gradually becomes an important part of network active security defense.The traditional Intrusion Detection System(IDS)has the problems of insufficient performance and efficiency.Therefore,one of hot research topics in the field of network security is to select the appropriate algorithm and design an efficient intrusion detection model.Aiming at the problems of the above intrusion detection methods,first,a model of SVM feature selection and parameter optimization is proposed based on improved fireworks algorithm.Support Vector Machine(SVM)is a machine learning algorithm that realizes the real risk minimization by seeking to minimize the structural risk.The goal of SVM is to pursue the optimal result under the condition of limited information.It constructs nonlinear mapping model and effectively solves the problem of high-dimensional,nonlinear and small samples.However,its model parameters(such as penalty term,kernel function parameters)and feature selection have a great impact on the classification performance,and many researches are based on the singularity of feature selection algorithm or SVM parameter optimization to improve the classification performance,but not Considering the intrinsic relationship between feature subset and support vector machine,to a certain extent,the detection effect is limited.In this thesis,we combine the 0-1 features of the feature selection problem with the improved binary fireworks algorithm to search the combination of feature data sets and SVM parameters.Combining the feature sets and SVM parameters as the binary fireworks algorithm,the combinatorial optimization strategy ensures that the classification accuracy is improved while selecting as few features as possible.Finally,based on the validity of the above algorithm by UCI data simulation experiments,an intrusion detection model is reconstructs based on the improved fireworks algorithm and SVM fusion.In the process of constructing intrusion detection model,this thesis effectively process some discrete,continuous and string data in KDD99 samples through numerical and normalized operations in the data preprocessing stage.After that,using the detection rate,false positive rate and correlation coefficient as evaluation criteria of the model performance,the model proposed in this thesis is compared with those of optimizing only SVM parameters,joint optimization of feature subsets and SVM parameters and stepwise optimization of feature subsets and support vectors Machine parameters three similar network intrusion detection model for performance comparison,the results show that the proposed intrusion detection model has better detection performance and good learning ability.