Decision-tree-based Support Vector Machines Of Removing Noise With Fuzzy Membership

Using support vector machine algorithm to solve two classification problems already has a very mature development, but the algorithm itself is very sensitive to noise. And for multi-class classification problems, it still needs further research and improvement. Especially in hierarchical Support Vector Machines algorithm, there will be the error accumulation problem.In response to these circumstances, this paper presents a new method to solve multi-class classification problem:adopt fuzzy membership to remove noise, generate support vector machine decision tree based on inter-class reparability. Compared with traditional methods, the model in this article avoids the phenomenon of refusal to partition, achieve the purpose of reducing the accumulated error, and the classification performance is also improved. This paper shows the feasibility of this method from the two aspects of mathematics theory and experiment result.The advantages of the method proposed in this paper is mainly reflected in:1, reducing the training sample set, reducing the effect of the noise points to the selection of the optimal classification hyper plane, shortening the training time; 2, using the decision tree support vector machines:1) not have to traverse all the sub-classifier when testing, saving time; 2) avoid the emergence of the refusal region, and the situation that the sample belongs to multiple classes or does not belong to any one kind; 3) The bottom-up clustering approach to obtain a more reasonable classification model, as well as good generalization ability.