A General Fuzzy Min Max Neural Network with Compensatory Neuron Architecture.

This paper proposes "A General Fuzzy Min-max neural network with Compensatory Neurons architecture"(GFMCN) for pattern classification and clustering. The network is capable of handling labeled and unlabeled data simultaneously, on-line. The concept of compensatory neurons is inspired from reflex system of the human brain. Fuzzy min-max neural network based architectures use fuzzy hyperbox sets to represent the data cluster or classes. An important stage in the training phase of these techniques is to manage the hyperbox overlaps and containments. In case of GFMCN, compensatory neurons are trained to handle the hyperbox overlap and containment. Inclusion of these neurons with a new learning approach has improved the performance significantly for labeled as well as unlabeled data. Moreover accuracy is almost independent of the maximum hyperbox size. The advantage of GFMCN is that it can learn data in a single pass (on-line). The performance of GFMCN is compared with "General Fuzzy Min-max neural network" proposed by Gabrys and Bargiela on several datasets. [ABSTRACT FROM AUTHOR]