An integrated mixture of local experts model for demand forecasting

Demand forecasting is an important issue in the supply chain planning process due to its usage for efficiently managing different planning tasks. The common techniques adopted to perform such task are easy to use forecasting methods performed in computerised systems with minimum human intervention such as the simple exponential smoothing. Concerning the drawbacks of these modelling approaches, past evidence suggests that they have been less successful in predicting short-term fluctuations on demand. In this work, the simple exponential smoothing method is employed to build an integrated mixture of local experts model (IMLEM) for demand forecasting using as independent variable a lagged gross domestic product (GDP) indicator. The IMLEM is usually employed to deal with rapidly changing situations and it is used in this work to improve the performance of the forecasts by providing a way to take into account short-term fluctuations on demand, due to changes on economic activity growth, within an easy to use forecasting method. The proposed method was applied on one-step ahead demand forecasting for a Brazilian beverage manufacturer for 130 non-seasonal stock keeping units (SKUs) and the accuracy of the generated forecasts allowed to reduce the forecasting errors. The obtained results also supported the decision making process by aiding demand planners concerning how to react to short-term fluctuations on demand.