Incremental Learning Algorithm Of Support Vector Machine Based On Vector Projection

Support vector machine (SVM) proposed by Vapnik and some other scholars is one of the standard tools for machine learning. Based on the statistical learning theory and optimization theory, it is an implementation of structure risk minimization principle in the statistical learning theory. But as a new technology developing, SVM, which still has some limitations, needs to be further explored and perfected.The researches included in the thesis can be summarized as follows:The case that non-support vector turn to support vector not to be considered in the basic incremental algorithm of Support Vector Machine. For this situation, an improved incremental algorithm is proposed; it makes use of the condition of KKT to deal with the history simples efficiently. This algorithm can improve the classifying accuracy.The concept of pre-extracting is introduced into Support Vector Machine. A new Incremental learning algorithm of support vector machine based on vector projection is proposed finally. In the algorithm, the thinking of incremental learning combine with pre-extracting avoiding pre-extrcting disabled. The geometric character of support vector is used to extract the simples which can be the support vectors possibly from incremental simples. The method of vector projection is used to generate the bound support vector set. In the incremental learning process, the bound support vector set as the training set which greatly reduces the number of training data in incremental SVM training, speeding up the training process. The experiment has been made on University of California-Irvine database. The result shows that the speed of training is improved remarkably and it is good solution to solve Support Vector Machine problems that need to deal with huge data set.