Network Application Layer Fault Diagnosis Based On Support Vector Machines

Support vector machine is a new machine-learning algorithm based on the Statistical Learning Theory, which has a fine performance with the limited samples. The SVM not only controls the complexity of the machine learning, but also has nice convergence speed and generalizing ability. The SVM transforms machine learning to solve a quadratic problem so that all the global optimal solutions can be found in the theory and conquer the "Course Dimensionality" with the kernel function. It has become an increasingly popular technique in machine learning.As fault diagnosis is a limited sample subject, we can use the SVM method to solve the problem of the Fault Diagnosis in the network application layer. The main research work in this paper is as follows:(1) A detailed analysis about the theory of the SVM is given in the paper. And one framework of the Network Fault Diagnosis System based on SVM in network Application Layer is proposed. The handle process and the function, mechanism of the components of this framework are discussed in the paper.(2) When the size of the training sets is uneven, the classification error rate of the results based on the traditional C-Support vector machine is undesirably biased with the less sample in the training sets. The reason of this phenomenon is discussed in the paper, and a method of using different weights of the penalty parameter for each classes is used to reduce this undesirable effect.(3) One SVM algorithm which to solve the problem when input samples make the different contributions to the decision surface is discussed in this paper. And one method is presented to give each sample different weight of the penalty parameter by using a fuzzy membership. This method can be used to reduce the effects of outliers and the noises, and improve the accurate precision of the SVM.