A new classification algorithm research

Classification hypersurface plays a very important role in classification problem. In SVM (support vector machine), classification hypersurface is emphasized because of the direct induction of the support vectors. In this paper, a measure of determining the importance level of the attributes based on classification hypersurface acquired by SVM is proposed. In traditional SVM solution algorithms, objective function is a strictly convex unconstrained optimization problem, but it is undifferentiable due to x+ . Therefore, the most used optimization algorithms are precluded. This paper presents a new technology to approximate the original undifferentiable model, so that the traditional SVM model is converted into a differentiable model. The proposed approach is experimentally validated in the datasets that are benchmarks for data mining applications.