A Rough Set-Based Clustering Method with Modification of Equivalence Relations
説明
This paper presents a clustering method for nominal and numerical data based on rough set theory. We represent relative similarity between objects as a weighted sum of two types of distances: the Hamming distance for nominal data and the Mahalanobis distance for numerical data. On assigning initial equivalence relations to every object, modification of slightly different equivalence relations is performed to suppress excessive generation of categories. The optimal clustering result can be obtained by evaluating the cluster validity over all clusters generated with various values of similarity thresholds. After classification has been performed, features of each class are extracted based on the concept of value reduct. Experimental results on artificial data and amino acid data show that this method can deal well with both types of attributes.
