A multiple classification method based on the D-S evidence theory

Based on D-S evidence theory which can handle uncertain information, a method for high-dimensional multiple classification problems was proposed. This method transformed high-dimensional multiple classification problem into several low-dimensional classification problems and established the classification support degrees of training samples in low-dimensional space. The low-dimensional classification support degrees of test samples were calculated by using k-nearest neighbour method, and they were fused by D-S evidence theory to obtain the classification support degrees in high-dimensional space. In order to decrease the effect of noise samples and the dispersion of samples, the Range Correction Coefficient and Inner-class Correction Coefficient were proposed. Several data sets were selected for comparative experiments, and the results show that the proposed method is more accurate and stable, especially for high-dimensional multiple classification problems.