Product demand forecasting with a novel fuzzy CMAC

Forecasting product demand has always been a crucial challenge for managers as they play an important role in making many business critical decisions such as production and inventory planning. These decisions are instrumental in meeting customer demand and ensuring the survival of the organization. This paper introduces a novel Fuzzy Cerebellar-Model-Articulation-Controller (FCMAC) with a Truth Value Restriction (TVR) inference scheme for time-series forecasting and investigates its performance in comparison to established techniques such as the Single Exponential Smoothing, Holt's Linear Trend, Holt-Winter's Additive methods, the Box-Jenkin's ARIMA model, radial basis function networks, and multi-layer perceptrons. Our experiments are conducted on the product demand data from the M3 Competition and the US Census Bureau. The results reveal that the FCMAC model yields lower errors for these data sets. The conditions under which the FCMAC model emerged significantly superior are discussed.