Twin Binary Tree Support Vector Machine Classifiers

Support vector machine on the basis of statistical learning theory by Vapnik et al. De-veloped a new type of machine learning methods, is a powerful tool for a solution to the small sample size problem, and has been widely used in many fields.This paper focuses on the twin binary tree support vector classifiers related issues, mainly for the following research:In chapter one, a brief introduction to the research background of the support vector machine, and from a new support vector machine model, the training speed and classifica-tion algorithm, these three aspects elaborated the current research status of support vector machine.In chapter two, we introduced the optimal separating hyperplane,(nonlinear) linear support vector machine and linear (nonlinear) twin support vector machine theory and related theory.In chapter three, for the binary tree support vector machine classification shortcomings, we consider the influence of the degree of separation between classes, combined with complete binary tree and partial binary tree, with using twin support vector machine in the decision node of binary tree to train got BT-TSVM and PBT-TSVM two kinds of classification al-gorithm, through the time complexity analysis BT-TSVM and PBT TSVM two algorithm are better than OVA-SVM, and by introducing a method of coordinate rotation and shrink-ing technology, thus obtained the CCBT-TSVM and CCPBT-TSVM two fast classification algorithm.In chapter four, according to the algorithms puts forward in the third chapter, for the artificial dataset, analysis can be obtained from the results of non-linear classifier, in data set2CCBT-TSVM not only has the highest classification accuracy rate of95%, and training speed than BT-TSVM shortened by5%,94%shorter than the OVA-TSVM; for real data sets, on the Ecoli, CCPBT-TSVM classification accuracy than OVA-SVM7%,3%higher than the BT-TSVM, training faster than OVA-SVM97%shorter than the BT-TSVM shortened by nearly1%, on the Glass, CCBT-TSVM has the highest classification accuracy, faster than93%OVA-SVM in the training speed. numerical experiments show that:our algorithm overall performance is better than the OVA-SVM, especially in dealing with large-scale and sparse strong data, CCBT-TSVM and CCPBT-TSVM these two kinds of algorithm of time advantage is more obvious.In chapter five, we summarize the main content of this paper and relevant conclusions for the future research work further outlook.