Classification and Prediction Method of Traditional Chinese Medicine Syndrome Based on Multi-label Learning and Bayesian Network

Abstract

The invention relates to a traditional Chinese medicine syndrome type classified predication method based on multi-label learning and Bayesian network. By finding a relation of six syndrome types of traditional Chinese medicine diabetes mellitus, an invisible formation cause of each syndrome type is excavated and the formation causes are combined with information from four diagnostic methods, so as to construct an augmentation characteristic set for describing a sample. Finally, a classifier is constructed through a feature selection algorithm and a multi-label classification algorithm, so that classified predication of six common syndrome types of the traditional Chinese medicine diabetes mellitus is realized.

Description

technical field [0001] The present invention relates to a method for predicting information classification, in particular to a multi-label learning-based TCM clinical symptom classification that uses label correlation as a supplementary feature to describe samples and combines multi-label learning algorithms with Bayesian networks method. Background technique [0002] The classification of TCM clinical symptoms is mainly to obtain the patient's symptom information (such as: headache, cold limbs, thready pulse, etc.) . The biggest characteristic of TCM clinical syndrome type classification is that there are often multiple corresponding symptom types for each patient, for example: deficiency of both Qi and Yin combined with blood stasis, which includes three syndrome types of Qi deficiency, Yin deficiency and blood stasis, so multiple markers are used Learning a model to build a classifier has become a common way to solve this problem. Based on the different ways of examini...

AI Technical Summary

Problems solved by technology

Based on the different ways of examining the correlation between labels, the existing multi-label learning problem solving strategies can be roughly divided into three types: first-order, second-order and high-order; the first-order method transforms the multi-label problem into multiple independent For binary classification problems, the relationship between labels is ignored, so the generalization ability of the model is the lowest, and the accuracy rate is not high; the second-order method splits the multi-label problem into pairwise label comparisons, which improves the generalization ability and correctness of the classifier to a certain extent. rate, but when the real problem has a correlation beyond the second order, the performance of this type of method will be greatly affected; the high-order method strategy constructs a classifier by examining the high-order tag correlation, such as processing any tag pair Influenced by all other markers, this type of method tends to have the highest generalization ability, but its complexity may also increase accordingly, which is not conducive to processing large-scale data

Images