Method, system and device for improving classification precision of support vector machine, and medium

Abstract

The invention discloses a method, system and device for improving the classification precision of a support vector machine, and a storage medium, and the method comprises the steps: extracting samplefeature data, and constructing an SVM classification model optimization problem based on the sample feature data; selecting two working variables which violate the SVM model optimization problem and have the most serious KKT conditions based on boundary limitation; updating the two working variables according to a quadratic programming solving method and carrying out next selection until all the working variables meet a KKT condition; obtaining parameters of an optimal SVM classification model based on all the updated working variables, and classifying the to-be-tested data in the support vector machine according to the parameters. According to the method, the strategy of selecting the working variables is optimized through boundary limitation, and the problems of low precision and poor performance caused by no progress of optimization are prevented; and only the parameter variation is calculated, so that unnecessary operation is simplified, the iteration efficiency of the SMO algorithm is greatly improved, and the classification precision is improved.

Description

technical field [0001] The present invention relates to the field of support vector machines, and more specifically refers to a method, system, computer equipment and readable medium for improving classification accuracy of support vector machines. Background technique [0002] After theoretical derivation, the optimization problem of the support vector machine learning model is finally transformed into a convex quadratic programming optimization problem in which the variable size is proportional to the number of samples. There are many methods for solving such problems, and SMO (Sequential Minimal Optimization) is widely used at present. , sequence minimization optimization) method. The SMO algorithm reduces the scale of the problem to two variables, and then converts the convex quadratic programming problem into a single variable convex optimization problem, which greatly reduces the complexity of the problem. [0003] The main advantage of the SMO algorithm is that it re...

AI Technical Summary

Problems solved by technology

[0004] However, in the process of actual use, it is found that in the process of selecting working variables, some situations where the optimal solution is not obtained but the parameters stop updating often occur, which causes the algorithm to fall into infinite meaningless iterations, and the algorithm performance is low and the optimal solution cannot be obtained. untie

Images