To accelerate the training accuracy and speed of the smooth support vector machine, the fuzzy clustering smooth support vector machine(FCSSVM) is proposed. The fuzzy clustering means(FCM) is used to decompose the training data into several sub-clusters, the entropy function is introduced to approximate the plus function of the slack vector, and an exact smooth model is derived buy using the expression of the weight vector at the optimal solution. The nearest neighbor subspaces of the test sample are defined, and the selective ensemble strategy is constructed to combine the classification results of several nearest neighbor subspaces. Numerical experiments demonstrate that FCSSVM has high accuracy, little iteration, good robustness and low training time.