Study Of Support Vectors Machine Algorithm Based On Pre-extracting Boundary Vectors

By using the statistical learning theory, support vector machine (SVM) is thought of a new generation of learning machine. It has been successfully applied into the field of data classification and regression such as handwritten digit recognition, face recognition, text classification, regression prediction and time serial analysis etc. The compute complexity of SVM is determined by number of training sample set. While many questions of reality have large amount of training sample set, it makes SVM cost many time to train.Based on SVM theory, the optimization separating hyper-plane just is determined by support vector. But mostly it is a small part of training sample set. Therefore if we can select a smaller samples set before training, which includes all support vectors, it will reduce much time while we use this set to train SVM.Based on many algorithms about pre-extracting boundary vectors this thesis include algorithm of pre-extract support vector into two class. One is algorithm of pre-extract boundary vector based on category centers, another is algorithm of pre-extract boundary vector based on NN.But the first one only apply to pre-extract boundary vector from samples set distribute euqality,thus this thesis proposed a mathematic model to improve it. The second one pre-extract boundary vector set is often too big or small, so this thesis proposed a new algorithm by using Density-Based Clustering to improve it. The experiment proved that using improved algorithms didn't decrease prediction precision. It had good result of improving pre-extract boundary vector set and reducing time of training.