The Structure Analysis And Application Of The Clustering Validity Index

As an important branch in the area of data mining, clustering technology aims to employ certain method(algorithm) to partition any data set into meaningful classes(or clusters), so that samples(data) within a cluster are similar in a certain standard while samples in different clusters show higher differences. Clustering has a wide range of applications, such as the fields of pattern recognition?biological medicine?industrial production and so on.In the process of clustering, the design(selection) of the clustering algorithm and cluster validation are both very important. To solve different problems in specific fields, at present, different clustering algorithms have been proposed, but they all have local applicability and limitations. Cluster validation proposed a quantitative standard for a certain data set to evaluate the advantages and disadvantages of different algorithms or sensitivity of the same algorithm with different parameters.Based on the existing clustering algorithms and validity analysis, in this paper, the characteristics of spectral clustering developed in recent years was analyzed, and a new clustering validity index based on NJW(Ng-Jordan-Weiss) algorithm was proposed. The choice of clustering algorithms based on the cluster validation is proposed in the field of the electrical tomography technology.The main work done in this paper is summarized as follows:A brief description of the clustering algorithms and cluster validity indexes was provided, and the application background and limitations of some typical clustering algorithms were analyzed.The characteristics of spectral clustering algorithm was summarized, and then a new clustering validity index based on NJW algorithm was proposed, the index achieved well results both in the tested artificial data sets and UCI data sets.The clustering algorithms are applied to electrical tomography technology. Four different algorithms were selected to the sample data sets and the evaluation of the results based on the Silhouette index was performed, and the experimental results validate the selection for different clustering algorithms by the Silhouette index.