A New Method of Attribute Reduction Based on Gamma Coefficient

In this paper, nonparametric method in statistics is introduced to analyze the reduction of ordinal attributes. Firstly,Goodman-Kruskal gamma coefficient, which is used to measuring the correlativity between ordinal variables in statistics, is now used as a new measure of correlativity between ordinal attribute sets after modifying it properly. Then, based on this new measure, a new method of attribute reduction for ordered decision tables is presented, and some connections between the method and the rough set theory about attribute reduction can be found. Furthermore, the numerical experiments show that the nonparametric statistical method we proposed is feasible and efficient for both consistent and inconsistent ordered decision tables.