A Multi-class Classification Algorithm Based on Hypercube

Abstract

A multi-class classification algorithm based on hypercube is proposed. For each class of training samples, a minimum hypercube that surround all samples is constructed in sample space. If two hypercubes intersect, the hypercube centers are used as the benchmark for compression. For a sample to be classified, its class label is determined according to the hypercube in which it is located. If this sample is not in any hypercube, the distances from the sample to the center of each hypercube are calculated firstly, and then the class label is determined by the nearest neighbor rule. The experimental results show that the training speed and classification speed of the proposed algorithm are improved significantly while ensuring the classification accuracy, especially in the case of large dataset and large number of classes.