Weighted Clustering Approach Based on Rough Set Similarity Model

In the process of clustering analysis, property characteristics are not equally important in the clustering process, and some characteristics are even redundant. Improper selection of characteristics may worsen effects of several classification methods. Therefore, correct selection of characteristics of great importance for clustering is crucial in improving clustering effects. Based on this, this paper proposes a new weighted clustering approach based on rough set similarity models, as well as a class purification approach based on information theory. First, weights of each property characteristic are initialized to 1, form primary equivalent classes with rough set similarity models, and then evaluate these property characteristics according to the information entropy theory, repeated weighted clustering are performed on each property with their entropy values, and finally obtain pattern classes satisfying clustering requirements. Tested by dataset from UCI, this algorithm is found to have obvious improvements in both classification rates and purity of each pattern class.