Study And Improvement Of The Clustering Method Applying In The Constitutional Classification

The constitutional classification of traditional Chinese medicine (TCM), which is the core and difficult point of the constitutional research, has not been solved very well and is hard to apply to the clinical practice. At present, there are lots of classifying methods for constitution; this severely restrains the development of TCM constitution.The clustering analysis classified without experiences, so it has scientificalness and objectivity. This paper analyzes the clustering methods systematically and entirely. In terms of the sample character, this paper firstly discusses the application of the clustering method in the constitutional classification of TCM, such as systemic method and k-means method, which is sensible to initial partitions (values of K) and initial cluster centers. So, it is hard to satisfy the need of clustering. In order to deal with these defects, this paper proposed a modified dynamic clustering Abstractmethod associated with the need of TCM classification. Be different from the method before, this algorithm chose some dispersive initial cluster center according to distance; meanwhile, it filter noise data in order to improve the clustering quality; in the course of clustering, the cluster radius and results values are changed dynamically, thus the more reasonable result is obtained. In the end, we apply the improved algorithm to analyze the sample data of the TCM constitution, and use the TCM knowledge to explain it. The clustering method provides an effective scientific proof for the constitutional classification of TCM.To promote the application of clustering analysis in TCM constitutional research, more emphasis should be laid on the decision of the number of cluster categories and explanation of cluster result. In order to make classification more reasonable, we should consider the priority of each character of sample data. Both of these are the following research emphases of this paper.