An improved approach to feature selection

Abstract

The paper addresses the feature selection based on Neighborhood Rough Set (NRS) used as evaluation function and Ant Colony Optimization (ACO) as generation procedure. A NRS-based measure is employed as heuristic information of ACO. For the weakness of setting a specified value to the size of neighborhood, a new standard deviation based value is advanced to be the size of neighborhood. Four datasets from UCI are used to evaluate the proposed approach and the experimental results show that the approach has a better performance, and could be a practical algorithm to select features from dataset.