A common attribute reduction form for information systems

An information system is an important form of knowledge representation, and attribute reduction plays an important role in machine learning, data mining, and intelligent systems. Several techniques are available to solve problems of attribute reduction but a common characterization for them is needed. This paper proposes the concepts of exact reductions and their reduction-invariant matrices. We obtain a unified mathematical model of attribute reduction by exactness for information systems, and show that frequently used methods of attribute reduction for information systems are exact. Specifically, we show that the positive region reduction for a decision table is exact. Our model theoretically unifies frequently used approaches to reduction. We also used a case study using the UCI dataset to verify the effectiveness of our proposed model. (c) 2020 Elsevier B.V. All rights reserved.