Simulation input data modeling

Input data modeling is a critical component of a successful simulation application. A perspective of the area is given with an emphasis on available probability distributions as models, estimation methods, model selection and discrimination, and goodness of fit. Three specific distribution classes (lambda, SB, TES processes) are discussed in some detail to illustrate characteristics that favor input models. Regarding estimation, the authors argue for maximum likelihood estimation over method of moments and other matching schemes due to intrinsic superior properties (presuming a specific model) and the capability of accommodating messy data types. They conclude with a list of specific research problems and areas warranting additional attention.