A feature saliency measure in WFMM neural network-based pattern classification

In this paper we present a feature saliency measure for the pattern classification based on FMM neural network. For the purpose we first introduce a modified structure of the FMM neural network in which the weight concept is added to represent frequency factor of feature values. In the model a hyperbox can be expanded without considering the hyperbox contraction process as well as the overlapping tests. During the learning process, the feature distribution information is utilized to compensate the hyperbox distortion which may be caused by eliminating the overlapping area of hyperboxes in the contraction process. The weight updating scheme and the hyperbox expansion algorithm for the learning process are also described. The proposed model can be applied to feature analysis process in a pattern classification problem. We define feature saliency measures that represent a degree of relevance of a feature in a classification problem. We consider three types of saliency measures: the relevance relationships between a feature value and a hyperbox, between a feature and a class, and the relevance of a feature in a given problem. The experiment results using the proposed model are also discussed.