Alternative output representation schemes affect learning and generalization of back-propagation ANNs; a decision support application

摘要

Inherently fuzzy outputs of an Artificial Neural Network (ANN) can be represented by various ways such as, a set of output neurons, one discrete level output neuron, one continuous output neuror. The effect of the aforementioned alternative output representation schemes to performance measures of the ANN, namely convergence of the learning procedure and generalisation capability of the ANN, is studied in this paper, Comparative classification results are presented for a Decision Support Systems (DSS) application namely the estimation by a ANN trained with Back Propagation, of bull/bear situation of the Athens Stock Market, given the values of a set of technical indicators.