| The automation of forecasting models selection process is inevitable for two reasons. First, the limited capability of human forecasters to choose the appropriate models. Second, the selection method itself is a cumbersome task.;Therefore, a new statistical approach is developed and evaluated, under a variety of conditions, using simulated time-series data. The purpose of this approach is to identify the pattern in the data history to be used in generating forecasts. The major patterns considered in this study are horizontal seasonal, multiplicative trend seasonal, additive trend seasonal, linear trend, damped trend, exponential trend, and horizontal pattern. The approach is then integrated with expert system forming a framework for expert system for forecasting. The objective of this framework is to show the necessary components and identify the relationships between them that facilitate the selection of appropriate forecasting model(s).;A prototype expert system, using an expert system shell, is built to show the feasibility of such a framework and to demonstrate its operations. A remarkable success, as shown by the prototype, is relieving forecast users from examining a great number of models in order to achieve the best possible model(s). |
