Based On The Research And Application Of Classification Of Incremental SVM Learning Algorithm

SVM(Support vector maehine)as a new statistical learning algorithm,with its outstanding theoretical basis(minimum structure theory,nuclear space theory).It is based on statistical learning theory developed a kind of general learning machine,its key idea is to use kernel functions to classify a complex task by kernel function mapping make it into a structure in high dimensional feature space linear classification hyperplane.Support vector machine(SVM)because of its excellent learning performance,has been widely used in the classification problem.Incremental learning is a kind of widely used technology of intelligent data mining and knowledge discovery technology,it is based on historical study results on the new data to study,makes the study has a certain continuity.In this paper,the main work is: first analyzes the theoretical basis,basic concept of the support vector machine(SVM),to solve the key technical problems and the basic concept of incremental learning.Then,this paper analyzes several existing incremental learning algorithm for support vector machines(SVMS),known from the analysis: most are not fully consider the new sample for initial sample concentration near support vectors of support vector,the influence of the useful historical data to be eliminated early,thus seriously affect the accuracy of classification.By introducing boundary support vector,put forward a kind of incremental learning algorithm based on boundary support vector,the experimental results show that the incremental SVM learning algorithm based on boundary support vector on the training speed and precision are improved.In addition,in view of the support vector machine,this paper analyzes the multiple classification problems,focusing on multiple classification algorithm based on the structure of the super ball is analyzed,it is concluded that an improved multiple classification incremental learning algorithm.Finally,the new algorithm in theapplication of text classification to do the design,verify the feasibility of this algorithm in actual application.