Support vector machine is a machine learning method based on statistical learning theory,which effectively solves the learning problem when the sample data is insufficient and has unique advantages in dealing with small samples,high dimensionality and nonlinear problems,and is widely used in the fields of face recognition and medical images.Many studies have shown that the selection of kernel parameters and penalty coefficients have a great impact on the classification effect when SVM is used to deal with classification problems,so how to reasonably select parameters to improve the classification accuracy has become a mainstream research direction in SVM.In this thesis,the Sparrow Search Algorithm is chosen to optimize the parameters of the support vector machine,but SSA is prone to fall into local extremes and the algorithm tends to converge prematurely.To address these problems,this thesis focuses on the improved Sparrow Search Algorithm and its parameter optimization for SVM.The specific work of the thesis is as follows.1)An improved sparrow search algorithm with mixed firefly perturbation(LEFASSA)is proposed,which firstly generates the initial sparrow population with an elite Oppositionbased learning to improve the diversity of the initial population;then introduces a Levy flight strategy in the position update formulae of discoverers and investigators to expand the exploration range of the algorithm;and finally incorporates an expanded step size firefly algorithm in the late iteration of the algorithm to improve the sparrow algorithm late in the algorithm iterations to improve the accuracy of the search for superiority.In order to verify the effectiveness of the proposed algorithm,it is simulated with standard SSA,FA,PSO,ABC and other sparrow search algorithm variants on 10 test functions,and to verify the rationality of the proposed strategy,the sparrow algorithm is tested separately for individual strategies.2)An SVM classification model is established based on the improved sparrow search algorithm,and an optimized SVM model of the improved sparrow search algorithm with mixed firefly perturbation(LEFASSA-SVM)is obtained,which is validated on 17 UCI machine learning datasets.In order to increase the conviction of the experiments,the proposed model is compared with several excellent swarm intelligence optimization algorithms,including GA-SVM,PSO-SVM,SSA-SVM and FA-SVM,and finally compared with the SVM classification results using linear kernel function linear,polynomial kernel function poly,Gaussian radial basis kernel function RBF,and hyperbolic tangent kernel function sigmoid respectively,and the experimental results show that the LEFASSA-SVM model can obtain higher classification accuracy and perform the best. |