FUZZY VERSIONS OF KOHONENS NET AND MLP-BASED CLASSIFICATION - PERFORMANCE EVALUATION FOR CERTAIN NONCONVEX DECISION REGIONS

Classification of certain linearly nonseparable pattern classes with n onconvex decision regions is a problem that cannot be efficiently hand led by the Bayes' classifier for normal distributions or other metric- based methods. An attempt is made here to demonstrate the ability of f uzzy versions of Kohonen's net and the multilayer perceptron for class ification of such patterns. In these models, the uncertainties involve d in the input description and output decision have been taken care of by the concept of fuzzy sets whereas the neural net theory helps to g enerate the required concave and/or disconnected decision regions. Sup eriority of these fuzzy models (over the respective conventional versi ons, the Bayes' classifier and seven other existing neural algorithms) has been adequately established when they are implemented on differen t sets of linearly nonseparable pattern classes. The effect of fuzzifi cation at the input has been investigated for both models. The contrib ution of the a priori probabilities of the pattern classes in the back -propagation procedure for weight updating has also been studied.