NEIGHBORHOOD ROUGH SET BASED ATTRIBUTE IN THE DECISION TABLE

Abstract

Attribute reduction is an essential data preprocessing step widely applied in pattern recognition, recommender systems, and decision support. For attribute reduction with numeric decision tables according to the granular computing approach, the traditional information granular is often shaded or discretized to build measures to evaluate the importance of the attribute. In this paper, we propose a new attribute reduction method including the following steps: 1) discretize data to increase the smoothness of the data after discretization; 2) determine the dependency ratio of the decision attribute with the set of condition attributes; 3) define a reduct and calculate the importance of the attribute to build an attribute reduction algorithm. The experimental results on the sample datasets from UCI show that our proposed method is effective compared to the attribute reduction methods following the traditional fuzzy computing approach.