A connectionist model for selection of cases

Abstract

The present article describes a method of designing a connectionist model for selection of cases for decision-making problems. The notion of fuzzy similarity is used for selecting the same from overlapping regions. Cases are stored as network parameters. The architecture of the network is adaptively determined through growing and pruning of hidden nodes under supervised training. The effectiveness of the cases, thus selected by the network, is demonstrated for pattern classification problem using 1-NN rule with the cases as the prototypes. Results, along with comparisons, are presented for various artificial and real life data for different parameter values of the similarity function, controlling the number of cases.