Sparseness Lssvm And Apply In Large-Scale Network Intrusion Detection Dataset

Least squares support vector machine is improved for the support vector machine. The problem of training stage is converted from quadratic programming to the linear equation solving, which reduces the computational complexity. But in the process, since the support values are not zero, the solution loses its sparseness and has effect on the performance of the vector machine.In order to achieve the purpose of sparseness optimization, this paper proposed the clustering algorithm for the selection of training samples, keeping data with high contribution value and remove the lower ones, which is the equivalent of setting the support value of corresponding portion to zero. Thus we obtain sparseness solutions and build the model simplely and effectively. For the choice of the clustering algorithm, this paper introduces kernel distance clustering and subspace PSO clustering algorithm which can effectively deal with high dimension and large data sets. Based on the kernel distance and cluster centers to construct sparseness support vector machine model which only contains valid data samples. Through the experiments on UCI datasets, although the two kinds of sparseness methods each has advantages and disadvantages, both can effectively dispose sparseness of the training sample. In the case of ensuring accuracy, they reduce the training phase consumption of resources, improve the efficiency of the model. This is a improve for the problem that LSSVM’s lacking of sparseness.At last, this paper apply the two kinds methods to the large scale intrusion detection data, the results of experiments further verifies the feasibility and validity of the sparseness methods.