Development of a neural network mathematical model for demand forecasting in fluctuating markets
Martin Ziarati, Erdem Bilgili, Lakhvir Singh, Başak Akdemir, Reza Ziarati · CERES (Cranfield University) · 2013
Research has shown that Neural Networks (NNs) when trained appropriately are the best forecasting system compared to conventional techniques. Research has shown that there is no system to accurately forecast sudden changes in demand for a given product. This paper reports on the development of a recovery method when a sudden change in demand has taken place. This error in forecasting demand leads to either excessive inventories of the product or shortages of it and can lead to substantial financial losses for the company producing or marketing the product. Two recovery methods have been developed and described in this paper: RZ recovery and Exponential Smoothing (ES). In the RZ recovery once a sudden change has taken place, a ‘soft’ Poke-Yoke (PY) system is setup warning the company that the normal forecasting system can no longer be relied upon and a recovery system needs to be initiated, with re-forecasting initiated.