CBCH (clustering-based convex hull) for reducing training time of support vector machine

摘要: Support vector machine (SVM) is an efficient learning technique widely applied to various classification problems due its robustness. However, the training time grows dramatically as number of data increases. As a result, applicability SVM large-scale datasets somewhat limited. In SVM, only few samples called support vectors (SVs) affect construction hyperplane. Therefore, removing irrelevant SVs does not degrade performance SVM. this paper clustering-based convex hull (CBCH) scheme introduced which allows efficiently remove insignificant and thereby reduce The CBCH initially applies k-mean clustering algorithm given points, then, each cluster obtained. Only vertices hulls points relevant are included points. Computer simulation over sizes types reveals that proposed considerably faster more accurate than existing classifiers. based on geometric interpretation applicable both linearly separable inseparable datasets.