A rough set based clustering algorithm and the information theoretical approach to refine clusters

In many clustering processes, the presence of more information does not usually generate a corresponding increase in the performance of clustering. The presence of irrelevant information decreases the effectiveness of the clustering algorithm. We propose a solution to improve the quality of clustering that is an attribute-weighted clustering algorithm based on rough set theory and the information theoretical refinement process. Firstly, we give every attribute the same weight value, and use rough set based clustering algorithm to get the initial classes. Then we weigh every attribute by Shannon's Entropy Theory, substitute mutual entropy values for the weight of every attribute, and compute with attribute-weighted rough set clustering algorithm again to refine and improve the clustering result. We have tested our algorithm on data sets from UCI repository. The experimental results show that our algorithm can obtain better results in classification rate and purity of classes than other traditional clustering methods.