Fuzzy cluster identification using neural networks

The paper deals with development of ARTMAP-like neural network to analyze feature space for classification purposes. The proposed tool is providing information about value of membership function of the unknown input vector to each class of interest. The designed ARTMAP-like system is called MF-ARTMAP based on the fact that membership functions are calculated. The functions shape is predefined as gaussian with adaptation of mean value and variance ill each feature space dimension during the training procedure. The usefulness of this approach is presented on the benchmark classification problems e.g. circle ill the square and spiral and on real-world data from satellite images over Slovakia. Results are similar when compared with Gaussian ARTMAP and provide additional information about membership function's values for tested input concerning classes of interest Classification accuracy is calculated using the contingency tables approach on actual and predicted classes of interest.