Nonnumeric Pattern Classifiers

Theoretical study was performed to determine if artificial neural systems can be used to generalize rules from examples. A method for generating rules from a multilayer network was investigated. When the network was presented with few training patterns, the rules derived from the network classified all patterns correctly. The generalization capabilities of the Pao-flu and neural network classifiers were compared with each other and with the ID3 method. Classifiers were compared by applying all methods to several data sets and examining the similarities and differences among them. It was demonstrated that the neural network could act as a rule generator for an expert system. The tests have shown that the network can correctly generate rules and subsequently correctly classify patterns.... Artificial neural networks, Pattern classifiers, Expert systems, Machine learning.