The Algorithm Research Of Support Vector Machine Based On The Decomposition Of The KD Tree

With the rapid development of information technology and spread of computerapplication, the increase in information and the spread of the data scale have reachedunprecedented growth speed. Finding useful information from big data has become a hot topicof the current machine learning.Support Vector Machines (SVM), which is proposed by Cortes and Vapnik, is the firststatistical learning theory based classification method, has achieved excellent both learningand generalization ability among machine learning algorithms. However, when the scale oftraining dataset is too large, the demand of computing resources raises too fast. In order toextend SVM to large scale dataset, this paper researches and analysises the issue from thefollowing aspects.Firstly, based on the ideas of local learning, this paper proposes a local classifierKDTSVM classification algorithm combining with KD tree and SVM classification algorithm.The algorithm partitions the training dataset into multiple local subspaces by using the KDtree structure and then constructs a local classifier for each local subspace using the SVMclassification algorithm.Secondly, for any testing sample, this article makes it traversing the KD tree until thetesting sample reaches a leaf node. If the class label of all the samples in the leaf node is thesame, the testing sample is divided into with these samples in the same class lable. Otherwise,using the local SVM classifier on the leaf node classifies the testing sample.Finally, this paper conducts extensive experiments on11sets of data set and shows thatKDTSVM algorithm improves the training and the testing speed of big data to some extentwhile keeping the testing accuracy, when it compares with DTSVM, LIBSVM, CVM.