Domain density description for multiclass pattern classification with reduced computational load

摘要

We propose a novel classification method that can reduce the computational cost of training and testing for multiclass problems. The proposed method uses the distance in feature space between a test sample and high-density region or domain that can be described by support vector learning. The proposed method shows faster training speed and has ability to represent the nonlinearity of data structure using a smaller percentage of available data sample than the existing methods for multiclass problems. To demonstrate the potential usefulness of the proposed approach, we evaluate the performance about artificial and actual data. Experimental results show that the proposed method has better accuracy and efficiency than the existing methods.