Stepwise selection of artificial neural network models for time series prediction

Abstract

Various heuristic approaches have been proposed to limit design complexity and computing time in artificial neural network modeling, parameterization, and selection for time series prediction. Nevertheless, no single approach demonstrates robust superiority on arbitrary datasets, causing additional decision problems and a trial-and-error approach to network modeling. To reflect this, we propose an extensive modeling approach exploiting available computational power to generate a multitude of models. This approach shifts the emphasis from evaluating different heuristic rules toward the valid and reliable selection of a single-network architecture from a population of models, a common problem domain in forecasting competitions in general and the evaluation of hybrid systems of computational intelligence versus conventional methods. Experimental predictions are computed for airline-passenger data using variants of a multilayer perceptron trained with backpropagation to minimize a Mean Squared Error objective function, deriving a robust selection rule for superior prediction results.