The Research Of Parameter Selection And Sparsity For Proximal Support Vector Machine

The solving of standard support vector machine is a quadratic programming problem. Itneeds long training time for large sample sets. In order to reduce the time complexity, manyscholars have put forth on proximal support vector machine. The standard proximal supportvector machine is a new type of solving classification problem, it is not a quadraticprogramming problem, but a regular least squares problem, and it can obtain the analyticalsolution, so it improves the training speed.This paper presents a proximal support vector machine based on the density-weightedmethod. It assigns a different degree for each sample, and it gives the density weight for thesample error according to density information of the sample. Proximal support vectormachine make all sample become support vector, so the sparsity can not control. This papergives a method of controlling the sparsity, and it effectively controls the sparsity. The qualityof the kernel parameter directly affect the generalization capability of the classifier, this paperpresents a kernel parameter selection method, so the classifier has better classificationperformance.This method is used in UCI data set, and it is compared with other methodsrespectively. The experimental results show that the incremental density weighted proximalsupport vector machines have better classification performance.