Research On Improved Algorithms Of Support Vector Machine

In this paper, we firstly introduced the background, research situation of support vector machines (SVM), and then review the basic theory of SVM. Based on these statements, further improvements are developed for least squares support machine algorithm and four classes classification problems.In this dissertation, the following are studied on basis of SVM:(1) Least squares support vector machine algorithm significantly improves the speed of large-scale training set and test speed, but it still has some shortcoming, such as how to dispose the training set with inconsistency between condition and decision attributes. We present an fuzzy rough set based least squares support machine (FRLSSVM) by considering the membership of every training sample, thus different training samples have a different contribution to the construction of the classification, at the same time, the consistence between the conditional features and decision features are taken into account.(2) We extended SVM to solving four class of classification problems, and proposed a new classification algorithm for four classes problems. This algorithm construct two hyperplane at the same time, the expression of this two hyperplane is solved at the same optimization problem. This algorithm has obvious advantages in reducing the number of classifiers and removing the unclassified regions.(3)Based on the four class of classification algorithm, furher study is made for 2k classification algorithm, in a similar accuracy, the 2k classification algorithm effectively improves the test speed, and removing the unclassified regions.To verify the effectiveness of these two algorithms, we both have done a lot of experiments for these two algorithm.