Investigation Of Parameters Optimization And Solution Method For Cost-sensitive Support Vector Machine And Its Application

Support Vector Machine(SVM)is a classification algorithm developed in the1990 s,which has been widely applied to pattern recognition,regression prediction and so on.Since traditional SVM is not cost-sensitive,it is possible that the negative sample can not be fully recognized in the practical application.For overcoming this difficulty,some researchers have constructed cost-sensitive SVM(CSSVM)recently.However,much time-cost is usually spent in the process of parameter optimization and solving of model when classifying large-scale data sets by CSSVM.This article will make some improvements for dealing with these problems.First of all,we propose a support vector pre-selection algorithm and a kernel parameter range pre-selection algorithm,and combine the two algorithms to design a CSSVM-ICC-DPRP algorithm for improving the efficiency of parameter optimization.The experiment of large-scale data sets shows that the CSSVM-ICC-DPRP algorithm can save more than 95% of the time compared with the grid algorithm,which shows the effectiveness of the algorithm.Secondly,for solving the CSSVM we transform it into an equilibrium problem,designs some algorithms,and proves the convergence of the algorithms.Then,the solution methods for the equilibrium problem are combined with the CSSVM-ICCDPRP algorithm,whose effectiveness is shown by data set experiments.Finally,by combining the index method which is offen used in the identification of violations with the algorithm in this paper,a feasible identification method is constructed to realize the real-time identification of the market power abuse in the spot market.Simulation experiments show that the identification method is effective.