Time series prediction for economic processes is a topic of increasing interest. In recent years artificial neural networks have been applied to this problem successfully [4], especially in the financial field. Neural networks can be used easier for the prediction of chaotic and noisy time series than statistical methods because they are able to learn the system dependencies on their own.
In order to reduce stock-keeping costs, a proper forecast of the demand in the future is necessary. In this paper we use feedforward multilayer perceptron networks for a short term forecast of the sale of articles in supermarkets. The nets are trained on the known sales volume of the past for a certain group of related products. Additional information like changing prices and advertising campaigns are also given to the net to improve the prediction quality. The net is trained by the back-propagation algorithm [2] on a window of inputs describing a fixed set of recent past states. For enhancement the algorithm is implemented on parallel systems.