EXPERIMENTAL COMPARISON OF TWO FEATURE SELECTION METHODS BASED ON GENERIC ALGORITHM

Feature selection is an important preprocessing in data mining, it aims to reduce the computational complexity of learning algorithm, and to improve the performance of data mining algorithms by removing irrelevant and redundant features. In the framework of discrete-valued feature selection, this paper experimentally compares two feature selection methods which are based on generic algorithm. The former uses relative classification information entropy to measure the significance of the candidate feature subsets, while the later uses inconsistency ratio to evaluate feature subsets. Both methods use genetic algorithm to search the optimal feature subset, we find by experimental comparisons that the former outperforms the latter.