Sample Clustering for Fast Classification by Using the Mean Shift Procedure

Most classification methods are limited by speed particularly when the training data set is large, such as artificial neural networks (ANNs) and support vector machines (SVMs). In this article, we explore the possibility of utilizing the mean shift algorithm, which is a mode seeking procedure that estimates the gradient of the data density, to decrease the sample size. We found that in a large number of samples to be trained, most samples can be clustered into a small number of mode centroids (extreme values of density), therefore, the original samples can be reduced by means of using the results of the mean shift procedure. To verify the validity of this method, several classifiers including the linear discriminant analysis (LDA), k nearest neighbor (kNN) and SVMs have been tested. Experimental results prove that when the parameters are selected appropriately, the proposed method is capable of reducing the computational complexity of above classification methods, with minimum effects on the classification accuracy.