Multiple Smooth Support Vector Machine With Fcm Clustering In Hidden Space

A multiple smooth model is proposed by smoothing technique and piecewise technique for large scale data. Mapping the training data to the hidden space with a hidden function, the proposed model divides the original data into several subclasses by Fuzzy C Means (FCM), whose initial cluster centers are selected by samples with large density indexes; derives the smooth differentiable model by utilizing the entropy function to replace the plus function of the slack vector, and introduces linking rules to combine results of subclasses. Simulations demonstrate that the obtained algorithm maintains good classification accuracies, reduces the training time and hardly varies with kernel parameters.