Incremental Learning Algorithm For Muti-class Support Vector Machine And Application In Cognitive Radio System

Support vector machine(SVM) is an pretty effective machine learning method, which performed well when dealing with non-linear finite sample. The key thought of the support vector machine is searching for an optimal super hyperplane in two categories of data to make the one class as far as possible away from the other one, and achieve good classification effect, thereby. Looking for the super hyperplane involve solving a quadratic programming problem(QPP), and it’s computational complexity is very large. In view of the multiple classification problems in everyday life, the use of traditional support vector machine classification method has exposed many malpractices, such as, classifier is non-parallel, low recognition rate, high computational complexity. Thus, multi-SVM were proposed. Truly, Muti-SVM algorithm has more wider application prospect.In real life, the training data will change constantly accumulated over time, so the traditional SVM algorithm has been unable to process the real-time data change, the incremental algorithm is proposed, therefor. In this paper, we propose an incremental learning algorithm based on Least square twin KSVC, which is the extended of TWSVM. Completing theoretical derivation and do some experiments. Do mathematical derivation in linear and non-linear, and choose sample data in UCI datasets, which has different classes to compare with incremental learning algorithm of classical SVM. Experiments show that the algorithm has advantages in:(1)traning speed;(2)classification rate;(3)dealing with muti-class data;(4)copying with high dimension data. Then we obtained the better recognition rate in communication system of signal modulation mode recognition problem of cognitive radio system.