Entropic feature discrimination ability for pattern classification based on neural IAL

Abstract

Incremental Attribute Learning (IAL) is a novel machine learning strategy, where features are gradually trained in one or more according to some orderings. In IAL, feature ordering is a special preprocessing. Apart from time-consuming contribution-based feature ordering methods, feature ordering also can be derived by filter criteria. In this paper, a novel criterion based on Discriminability, a distribution-based metric, and Entropy is presented to give ranks for feature ordering, which has been validated in both two-category and multivariable classification problems by neural networks. Final experimental results show that the new metric is not only applicable for IAL, but also able to obtain better performance in lower error rates. © 2012 Springer-Verlag.