Feature selection using neural networks with contribution measures

Abstract

There still seems to be a misapprehension that neural networks are capable of dealing with large amounts of noise and useless data. This is true to a certain extent but it is also true that the cleaner and more descriptive the data is the better the neural networks will perform, especially when dealing with small data sets. A method for determining how useful input features are in giving correct classifcations using neural networks is discussed here.