Novel Attribute Reduction on Decision Rules*

Abstract

From the perspective of formal concept analysis, the concepts of a formal context generated become larger in number with growing data. Attribute reduction based on decision formal context is to find out minimum subsets of attributes while maintaining the ability of classification, decision rules simplified as well which will make decision making much easier. This paper firstly generates decision rules, divides decision rules into strong rules and weak rules, puts forward judging theorems of non-redundant rules and rule reduction; secondly, proposes an approach of rule reduction by categories of attributes; in the end, discusses the time complexity. Comparing with other algorithms on runtime and ability of classification, experimental analysis shows that our method approves feasibility and accuracy. In the end, it draws a conclusion and discusses open issues.