Methodological Assessment for product demand forecasting through neural networks

In this paper we propose a new forecasting model featuring higher precision capabilities with respect to other methods available in literature. The proposed approach relies on the development of a computational technique based on the development of the Neural Network techniques, which are notoriously able to estimate the future product demand, especially when it is influenced by high instability and is dependent on exogenus factors. The forecasting delay, organised on the short-medium period, provides the future need for products spanning on a few months interval. The algorithm developed and the entire system realised has been tested with real historical data collected from three big firms operating in three different marketing environment. Finally, the results given out by our system have been compared with those obtained by traditional statistical methods. The best results have been achieved with some particular part/numbers, whose demand were characterised by a high degree of uncertainty. Our system showed an improvement of matching the forecasting data up to 25% compared with traditional methods and an improvement of the time to select the neural network parameters of up to 37%.