Research On Support Vector Machine Optimization Algorithm Based On Improved Multiverse Algorithm

The multiverse optimization is a new swarm intelligence optimization algorithm,the main inspiration comes from the big bang,with other existing group of intelligent optimization,there are also some shortcomings,is one of the most significant defects such as slow convergence speed,easily fall into local extremum.In recent years,some improved algorithms for defects have been put forward,but these improved algorithms only solve a small part of the problem,and do not integrate all defects to improve the algorithm.The same problem also exists in other similar swarm intelligence optimization algorithms.In recent years,in view of the support vector machine was optimized algorithm emerge in endlessly,swarm intelligence optimization algorithm,is part of a group of intelligent optimization algorithm can be used to support vector machine(SVM)classification model for performance tuning,and select the best parameters,including penalty function parameters C and kernel function parameters ?.The support vector machine model can map the original linearly indivisible data to a higher dimension space through the corresponding kernel function,so that the original linearly indivisible data can be linearly separable in the mapped space.Therefore,the kernel function parameter has a great influence on the classification performance.The penalty function parameters can balance the ability to fit the sample and the ability to predict the sample after testing.Thus,these two parameters on the performance of the model has a significant impact,how to choose the most appropriate parameter improve model performance is one of the important issues should be considered.To sum up,this paper proposes two improved multiverse optimization algorithms to solve the defects of the multiverse optimization algorithm,and uses the improved algorithm model to select the optimal feature subset and the most appropriate parameters to improve the performance and stability of the classification algorithm model.Firstly,an improved multiverse optimization algorithm model(COUMVO)is proposed,which uses logarithmic increasing WEP,compression factor and chaotic mapping to further solve the problems in multiverse optimization algorithm;Secondly,an improved multiverse optimization algorithm based on adaptive mechanism(BWTMVO)was proposed.BWTMVO algorithm balanced the two stages of swarm intelligence optimization algorithm through an improved multiverse algorithm based on adaptive mechanism,so as to improve the multiverse optimization model.The improved algorithm model is integrated with support vector machine to improve the classification accuracy of the model.In addition,we use an existing system architecture to describe the entire COUMVO-SVM classification model and the BWTMVO-SVM classification model execution process,and evaluate the performance of the two classification models on multiple standard UCI data sets.In this paper,the COUMVO algorithm and BWTMVO algorithm are compared and analyzed with standard MVO algorithm,several improved MVO algorithms and other multi-population intelligent optimization algorithms.Through the final experimental results,we can see that the COUMVO algorithm proposed in this paper has better performance than other algorithms of the same type in classification models on most data sets,and the second improved algorithm proposed in this paper,BWTMVO algorithm,has a more significant performance improvement compared with COUMVO algorithm and other multi-population intelligent optimization algorithms.