Optimal Selection of Left and Right Hand Multi-channel Pulse Features Based on Neighbourhood Component Analysis

The use of computer technology to analyze the pulse for disease diagnosis is an important research direction in the standardization of traditional Chinese medicine (TCM) diagnosis. TCM believes that the pulses at different positions of the left and right hands correspond to different organs. However, few studies have conducted pulse signal analysis based on this theory of TCM. Taking this theory into consideration, this research proposes a multi-channel feature selection method based on Neighbourhood Component Analysis(NCA), which uses multi-channel pulse features more comprehensively. First, select the most important features of each location through NCA, and construct multiple feature combinations. Then through the classification algorithm, the optimal feature combination under the corresponding algorithm is selected, and the feature subsets at different channels are constructed. The experimental results show that the left-hand three-channel can obtain better classification results by using pulse features of fewer channels, and has good performance, with a precision of 88.6% and a recall rate of 94.2%.