The comparison with improved mixture kernel SVM and traditional neural network

Support Vector Machines bases on statistical learning theory and replace the minimization experiential risk minimization by structural risk minimization, thus have large advantage over the traditional neural network on small sample set for classification. Related documents and experimental data prove that SVM is the best learning machine among all kinds recently and has large advantage over those of traditional neural networks. In this paper we prove that the performance of an improved SVM with mixed kernel will make the advantage more obviously. Different from some papers choose kernels and parameters randomly, we choose the kernels for SVM theoretically, through observing and computing the kernel matrix. Base on this, we used the selected kernel functions to get a new mixed kernel function. Experiential data proved that this new SVM has a better performance than that of that traditional neural network. This will give us a method to get a new learning machine for pattern identification.