An object oriented approach to forecasting

Abstract

This paper introduces the use of object oriented design techniques in the area of forecasting. It focuses on representing the knowledge underlying the forecasting process using classes, objects and properties organized in a tree structure. Three major classes have been identified: time-series, quantitative forecasting tools, and results. Objects constitute specific instances of these classes, while properties represent their characteristics perceived as decisive in the forecasting process. Reasoning is performed by rules that act upon the three basic structures of representation mentioned above and determine the inference relations that exist among them. In the working applications developed, knowledge bases corresponding to specific forecasting methodologies were integrated with the proposed object oriented representation.