A New Attribute Dependency Function in Information System

Attribute dependency function is very important for feature selection in data mining, pattern recognition and machine learning. However, Pawlak´s is inadequate for some information systems, and Daisuke´s definition is only for categorical attribute. In this paper, we introduce a new definition based on partition for numerical attribute. The advantage of the definition is that heterogeneous features can be dealt with. At last, we apply the function to local reduction, the experimental results show that the definition is more flexible to deal with heterogeneous features as a new quantitative analysis tool for local reduction.