抄録
This paper proposes a novel variable selection method based on the KL information measure, and applies it to optimal channel selection for bioelectric signal classification. Generally, the accuracy of classifcation for bioelectric signals is greatly influenced by measuring positions of the signals as well as individual physical abilities of a user. Therefore, it is effective for classification to select optimal positions for each user in advance. In the proposed method, the probability density functions (pdfs) of measured data are estimated through learning of a multidimensional probabilistic neural network (PNN) based on the KL information theory. Then, a partial KL information measure is newly defined to evaluate contribution of each dimension in the data. The effective dimensions can be selected eliminating ineffective ones based on the partial KL information in a one-by-one manner. In the experiments, the proposed method was applied to EMG electrode selection with six subjects (including an amputee), and the effective channels were selected from all channels attached to each subject's forearm. Experimental results showed that the number of channels was reduced with 36.1±12.5 [%], and the average classification rate using selected channels by the proposed method was 98.99±1.31 [%]. These results indicated that the proposed method is capable to select effective channels (optimal or semi-optimal) for accurate classification.
