Eliminating redundancy and irrelevance using a new MLP-based feature selection method

This paper presents a novel feature selection method based on the use of a multilayer perceptron (MLP). The algorithm identifies a subset of relevant, non-redundant attributes for supervised pattern classification by estimating the relative contribution of the input units (those representing the attributes) to the output neurons (those corresponding to the problem classes). The experimental results suggest that the proposed method works well on a variety of real-world domains.