Font Size: a A A

Triplet Support Vector Machines For Pattern Classification

Posted on:2011-02-04Degree:MasterType:Thesis
Country:ChinaCandidate:H P WangFull Text:PDF
GTID:2178360308480918Subject:Computational Mathematics
Abstract/Summary:PDF Full Text Request
As the most successful machine learning method during the last decades, Support Vector Machine(SVM) has been broadly applied in pattern recognition area and so on. The biggest difference from other machines learning methods is that SVM is corresponding to several principles in statistical learning theory. The support vector machine approach was originally developed to solve binary classification problems. How to effectively extend it for multi-class classification is still an on-going research issue.In this thesis, according to multi-class classification methods based on SVM and twin support vector machines, some problems of support vector machines algorithms are analyzed:(1) Aiming at the shortcoming of twin support vector machines, which can't solve three classification as well as multi-classification problem, a new triplet support vector machines based on the idea of twin support vector machines and one-against-rest support vector machine are proposed, and acquire very good results. After constructing the primitive problem of triplet support vector machines, the existence and uniqueness of the problem and the relationship with the dual problem are proved. The efficiency of the proposed methods is proved by results of experiment.(2) Least squares triplet support vector machines are also raised based on the idea of least squares SVM. It avoids the more onerous dual problem of the quadratic programming problem instead of solving three equations, and effectively improving the speed of a large sample study. After giving the original problem of least squares triplet support vector machines, its solutions are thoroughly analyzed. At last, good generalization capability of the proposed method is shown by experiments made on UCI datasets. The experiments are executed to compare on one-against-rest, one-against-one and the proposed methods, the result shows that least squares triplet support vector machines classifier is of the better performance.
Keywords/Search Tags:Statistical Learning Theory, SVM, Multi-class Problem, Twin Support Vector Machines, Least Squares SVM
PDF Full Text Request
Related items