Decision Table Reduction in KDD: Fuzzy Rough Based Approach

Abstract

Decision table reduction in KDD refers to the problem of selecting those input feature values that are most predictive of a given outcome by reducing a decision table like database from both vertical and horizontal directions. Fuzzy rough sets has been proven to be a useful tool of attribute reduction (i.e. reduce decision table from vertical direction). However, relatively less researches on decision table reduction using fuzzy rough sets has been performed. In this paper we focus on decision table reduction with fuzzy rough sets. First, we propose attribute-value reduction with fuzzy rough sets. The structure of the proposed value-reduction is then investigated by the approach of discernibility vector. Second, a rule covering system is described to reduce the valued-reduced decision table from horizontal direction. Finally, numerical example illustrates the proposed method of decision table reduction. The main contribution of this paper is that decision table reduction method is well combined with knowledge representation of fuzzy rough sets by fuzzy rough approximation value. The strict mathematical reasoning shows that the fuzzy rough approximation value is the reasonable criterion to keep the information invariant in the process of decision table reduction.