Identification of Feature-Salience

In this paper we present two techniques designed to identify the relative salience of features in a data-defined problem with respect to their ability to predict a category outcome-e.g., which features of a character contribute most to accurate prediction of outcome. The first technique we proposed is a neural-net based clamping technique and another is based on inductive learning algorithm-decision tree's heuristic. They are compared with a number of other techniques, i.e., automatic relevance determination (ARD), weight-product, random selection, in addition to a standard statistical technique-linear correlation analysis. The salience of the features that compose a proposed set is an important problem to solve efficiently and effectively not only for neural computing technology but also in order to provide a sound basis for any attempt to design an optimal computational system. The focus of this study is the efficiency as well as the effectiveness with which high-salience subsets of features can be identified in the context of ill-understood and potentially noisy real-world data