Establishment Of Bayesian Network Based On The Latent Structure And Its Application In Syndrome Differentiation Of TCM

With the rapid development of machine learning,Bayesian network has been applied to many fields,the Bayesian network has been a research hot spot of machine learning.The learning of Bayesian network with hidden variables is a important issue.These hidden variables can converge the complex dependencies between the observed variables,thus simplifying the network model.The learning of Bayesian network with hidden variables is a very meaningful research content.At the same time as the most basic elements of syndrome differentiation,Syndrome elements cannot be observed directly.Therefore,implicit variable has a very important role in TCM syndrome differentiation.In this paper,we first get the Bayesian networks with hidden variables by learning Bayesian networks,then apply it to TCM syndrome differentiation.Realizing the objective of TCM syndrome differentiation.The main work of this paper is the following two points:(1)Proposing a feature selection algorithm based on feature group selection.This algorithm first cluster the original feature and combined with the classification accuracy of the classifier to select the feature group.At the same time,a Bayesian network learning algorithm based on feature clustering is proposed.Applying the algorithm to the multi label learning of chronic gastritis data in traditional Chinese medicine and obtaining good classification results.(2)Proposing Bayesian Network classification algorithms with hidden variables which based on Multidimensional clustering(CBMC).This algorithm is mainly divided into two parts,which are unsupervised and supervised.Applying the algorithm to the multi label learning of chronic gastritis data in traditional Chinese medicine and obtaining good classification results which the Average Precision can reach 82.6%.