In this paper, we present a modified fuzzy min-max neural network
model and its application to feature analysis. In the model a hyperbox can be
expanded without considering the hyperbox contraction process as well as the
overlapping test. 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 described. A feature analysis technique for pattern classification using
the model is also presented. We define four kinds of relevance factors between
features and pattern classes to analyze the saliency of the features in the
learning data set.