Measuring the Quality of Semi-subjective Training Data and Its Application in the Inference of TCM Zhengs

This paper proposes a method for inferring a person's Zheng (evidence) status of traditional Chinese medicine (TCM) using the semi-subjective training patterns which are acquired from the multi-sensors including CCD camera, and semi-subjectively labelled by TCM doctors (TCMDs). After confirming the overlapping of samples belonging to the Zheng class and its negative class based on the extracted eigen attributes, an matric called qoSSD for measuring the quality of semi-subjective training data (QoSSD) used in the binary classification is defined, which are varied with the different Zhengs and TCMDs. On the basis of qoSSD, the sufficiency and effectiveness of the training data for inferring the Zhengs is estimated. The experiment verifies the positive correlation between qoSSD and the quality of the inference. Further, it confirms that the influence of modality change of sensors on the quality of the inference can be analyzed by QoSSD.