Automated Discovery of Partitions in Data Based on the Relations between Indiscernibility and Cluster Number

Authors: Shoji Hirano and Shusaku Tsumoto, Shimane University

This paper presents a new method for discovering partitions in data based on the relations between indicernibility and cluster number derived by rough clustering. Our previous the relationships between indiscernibility degree and the number of clusters draw a globally concave but multimodal curve, and the range of indiscernibility degree that yields best cluster validity appears as a local minimum around the global one which generates single cluster. Based on this feature, we attempt to automatically discover the value of indiscernibility degree that represents adequate granularity of data.