Economic prediction system using double models

This paper proposes a model-selection-based economic prediction system with double models. The two models reflect short term and middle-long term trends of economic series respectively. This prediction system can solve problems of overfit or underfit that may beset prediction systems with a single model. MEBML (mind-evolution-based machine learning) is a new method of evolutionary computation proposed by the authors (1998). MEBML has better convergence performance and computational efficiency than genetic algorithms. A prediction system using MEBML can determine the forms, the orders and optimum parameters of the models, as compared with other methods.