COMBINING EXPERT-SYSTEMS AND NEURAL NETWORKS FOR LEARNING SITE-SPECIFIC CONDITIONS

An artificial neural network to generate fertilizer recommendations fo r malting barley was developed. The neural network was trained and tes ted with data generated by a malting barley expert system developed by the authors and others. The purpose of creating a neural network was to explore the possibilities of using neural networks as tools to adju st knowledge based systems to site-specific conditions. It was shown t hat the neural network can adjust itself to site-specific conditions w ith fewer than 5% site-specific patterns in the training set. In addit ion the training set size required to enable the neural network to gen eralize was found to be at least 2000 patterns. (C) 1997 Elsevier Scie nce B.V.