Adaptive expansion algorithm for fuzzy min-max neural network in pattern classification

A fuzzy min-max (FMM) neural network classification model, classifies patterns using an hyperbox fuzzy sets, and the learning includes hyperbox expansion, overlap, and contraction steps. The maximum size of hyperbox controls expansion, which may lead to overlapping between hyperboxes of different classes and contraction step is used to get rid of overlapping hyperboxes. We propose an adaptive expansion algorithm for an expansion of hyperbox in FMM neural network for pattern classification. The expansion criterion of this algorithm calculates area covered by hyperbox. The area covered by hyperbox is a user defined value. The results show that numbers of hyperbox created are less and accuracy of testing is achieved at a minimum value of the area as correlated to the same size of hyperbox.