Abstract: Aiming at problems existing in the area of multi-class pattern recognition with large number of cata logs,a hypersphere support vector machine classification method based on semi-fuzzy kernel clustering is proposed.Samples are preprocessed with semi-fuzzy kernel clustering to ensure that the ones near boundaries are selected and then used to train hypersphere support vector machine so as to improve its performance efficiently.Some experimental results indicate that the new method yields higher precision and speed than classical support vector machine multi-class classification methods.

Key words: support vector machine, multi-class classification, semi-fuzzy kernel clustering, hypersp here