Multi-confidence rule acquisition oriented attribute reduction of covering decision systems via combinatorial optimization

Abstract

Rule acquisition is one of the most concerned issues in the study of decision systems including covering decision systems. Usually, a covering decision system is inconsistent, which can lead to the result that some of the rules derived from the system are not certain but possible rules. Considering the fact that, in addition to the certain rules, the possible rules with high confidence are also commonly used in practice for making decision, and the compact rules without redundant conditional attributes can conveniently be used by a decision maker, we propose in this study a rule confidence preserving attribute reduction approach in order to extract from a covering decision system both the compact certain rules and the compact possible rules with their confidence degree being not less than a pre-specified threshold value. Furthermore, a combinatorial optimization algorithm is formulated to compute all the reducts. Some numerical experiments are further conducted to evaluate the performance of the proposed reduction method.