Development and evaluation of a prototype expert system for forecasting models.

摘要

The selection of forecasting methods is a difficult task for which an expert system would offer an alternative if it could translate decision making procedures obtained from human forecasting experts into an explicit repeatable computer program. This research consisted of: (1) the development of a prototype expert forecasting system, dubbed the Intelligent Advisor for Selection of Forecasting Methods (IASFM), using a rule based shell with 199 production rules; and (2) an evaluation of the IASFM using the Analytic Hierarchy Process (AHP) developed by Thomas L. Saaty in the 1970s.; The following seven criteria recommended by previous researchers are used by the IASFM to select one or more of thirty-seven available forecasting methods: (1) time horizon, (2) data pattern, (3) cost, (4) ease of use, (5) model characteristics, (6) minimum data requirement, and (7) accuracy. The use of a rule based shell facilitated the development of the IASFM in a very short time period without the knowledge of an artificial intelligence language such as Lisp or Prolog.; The AHP was used to evaluate the IASFM because it is a decision analysis method for multiple criteria decision making. The following seven categories were used in the evaluation: (1) decisions, advice, and performance; (2) reasoning; (3) discourse; (4) hardware environment; (5) efficiency; (6) ability to update; and (7) ease of use. Three Mississippi State University statisticians were the evaluators. The consistency ratio for each was less than ten percent, which is the usual criterion for an AHP evaluation. The IASFM was recommended by two evaluators, while one evaluator recommended the use of a human expert.