Unvariate Time Series Forecasting with Fuzzy CMAC

Abstract

In financial and business areas, forecasting is a necessary tool that enables decision makers to predict changes in demands, plans and sales. This work applies a novel fuzzy cerebellar-model-articulation-controller (FCMAC) into univariate time-series forecasting and investigates its performance in comparison to established techniques such as single exponential smoothing, Holt's linear trend, Holt-Winter's additive and multiplicative methods and the Box-Jenkin's ARIMA model. Experimental results from the M3 competition data reveal that the FCMAC model yielded lower errors for certain data sets. The conditions under which the FCMAC model emerged superior are discussed.