A novel algorithm for feature selection based on rough set theory

This paper presents a forward-searching algorithm for feature selection, and applies it in classification problem. It adopt the number of the objects that can be correctly classified as the heuristic information and evaluation function, and it will stop until current optimal feature set is the same as that retrieved in previous step. This algorithm is implemented in 7 databases randomly selected from UCI. The result of the experiment shows that the feature set retrieved has the property of ldquonot decease classification accuracy obviously, not affect the distribution of class, stable and strong adaptabilityrdquo.