A Method for Improving SVM Classifier by Excluding Redundant Information

This paper proposes that support vectors include redundant information after analyzing Kernel´s geometrical structure and researching data dependant method for improving support vector machine (SVM). Redundant information confuses the law of a learning problem. Data dependant method on improving SVM is based on Riemannian geometry theory and could exclude redundant information. Reasoning and experiments show this method could effectively improve classification ability and classification speed of SVM.