Extension Model of Rough Set Under Incomplete Information

Yin Xu-Ri,Shang Lin

2006-01-01

Abstract:The classical rough set theory is based on complete information systems. The starting point of the rough sets theory is an observation that objects having the same description are indiscernible with respect to the available information. It classifies objects using upper-approximation and lower-approximation defined on an indiscernibility relation that is a kind of equivalent relation. But the indiscernibility relation may be too rigid in some situations. Therefore several generalizations of the rough sets theory have been proposed. Some of them extend the indiscernibility relation using more general similarity or tolerance relations. For example, Kryszkiewicz has studied the tolerance relation, and Stefanowski has explored the non-symmetric, similarity relation and valued tolerance relation. Unfortunately, these extensions have their own limitation. In this paper, several extension model of rough set under incomplete information are discussed. A concept of constrained similarity relation as a new extension of rough sets theory is introduced, and the upper-approximation and lower-approximation defined on constrained similarity relation are proposed. Furthermore, the performances of these extended relations are compared also. Analysis result shows that this relation can effectively process incomplete information and generate rational object classes.