Pattern recognition from neural network with functional dependency preprocessing

This paper describes how the functional dependency preprocessing technique can he used to enhance the performance of pattern recognition from a trained artificial neural network. By identifying the functional dependencies of a data set prior to network training, a subset of attributes of the data set can be found which can determine the classification attribute. Experimental results indicate that it can lead to faster network training, smaller neural network size and better (or at least equal) generalization accuracy of the network.