Short term prediction of sales in supermarkets

Abstract

In this paper artificial neural networks are applied to a short term forecast of the sale of articles in supermarkets. The times series of sales, prices and advertising campaigns are modelled to fit into feedforward multilayer perceptron networks that are trained by the back-propagation algorithm. Several net topologies and training parameters have been compared. For enhancement the back-propagation algorithm has been parallelized in different manners. One batch and two on-line training algorithms are implemented on parallel systems with both the runtime environments PARIX and PVM. The research will lead to a practical forecasting system for supermarkets.