Centroid Normal Direct Support Vector Machine

摘要

The present training method of support vector machine has large computational cost. Integrating SVM’s geometrical interpretation and structural risk minimization, a new fast support vector machine: Centroid Normal Direct Support Vector Machine (CNDSVM) was proposed. The normal vector of optimal separating hyperplane was determined by the connective direction of centroids. The optimal separating hyperplane was directly obtained through the distributing of the samples normal projection. The experiments on artificial data sets and benchmarks data sets show that the method can get almost the same classify result as SVM and LSSVM with very low computational cost.