Neuro-fuzzy pattern classification model with rule extraction based on supervised learning

The main objective of this paper is to design a new method for generating fuzzy rules for pattern classification. To start with, separation hyperplanes for classes are extracted from a trained neural network. The convex existence regions in the input space for each class is approximated by shifting these hyperplanes in parallel using the training data set for the classes. Using the fuzzy rules the numerical input data is classified directly without the need of neural networks. The proposed method is verified for target recognition using radar cross section signals.