A novel incremental attribute reduction approach for dynamic incomplete decision systems

Abstract

Attribute reduction is an important process in data mining and knowledge discovery. In dynamic data environments, the attribute reduction problem has three issues: variation of object sets, variation of attribute sets and variation of attribute values. For the first two issues, a few achievements have been made. For variation of the attribute values, current attribute reduction approaches are not efficient, because the method becomes a non-incremental or inefficient one in some cases. In order to address this, we first introduce the concept of an inconsistency degree in an incomplete decision system and prove that the attribute reduction based on the inconsistency degree is equivalent to that based on the positive region. Then, three update strategies of inconsistency degree for dynamic incomplete decision systems are provided. Finally, the framework of the incremental attribute reduction algorithm is proposed. Experiments on different data sets from UCI show the accuracy and feasibility of the proposed incremental reduction algorithms.