Selecting appropriate forecasting models using rule induction

Forecasting is a critical activity for numerous organizations. It is often costly and complex for reasons which include: a multiplicity of forecasting methods and possible combinations; the absence of an overall 'best' forecasting method; and the context-dependence of applicable methods, based on available models, data characteristics, and the environment. In recent years, artificial intelligence (AI)-based techniques have been developed to support various operations management activities. This research describes the use of one such AI technique, namely rule induction, to improve forecasting accuracy. Specifically, the proposed methodology involves 'training' a rule induction-based expert system (ES) with a set of time series data (the 'training' set). Inputs to the ES include selected time series features, and for each time series, the most accurate forecasting method from those available. Subsequently, the ES is used to recommend the most accurate forecasting method for a new set of time series (the 'testing' set). The results of this experiment, which appear promising, are presented, together with guidelines for the methodology's use. Its potential benefits include dramatic reductions in the effort and cost of forecasting; the provision of an expert 'assistant' for specialist forecasters; and increases in forecasting accuracy.