CHOOSING THE BEST MODEL - LEVEL OF DETAIL, COMPLEXITY, AND MODEL PERFORMANCE

The lack of a methodology, or at least detailed guidelines, for choosi ng the best model in a mathematical or computer modelling study stems from a poor understanding of the precise ways in which the success of the study depends upon the particular model used. As a result, the cho ice of the best model is regarded as more of an art than a science. In order to improve the model selection process, model performance needs to be clearly defined, and suitable model attributes identified that can be used to predict the performance of the alternative candidate mo dels. This paper distinguishes the different aspects of model performa nce and considers the extent to which they can be measured. The most c ommon attributes used to compare alternative models are level of detai l and complexity although these terms are used in a number of differen t ways. The meanings of these concepts are therefore discussed and the likely relationships with the model performance elements considered. The related area of simplification is reviewed and the areas in which further work is required are set out.