Smoothing time series for input and output analysis in system simulation experiments

Classical methods of studying the behavior of the output of a simulation model as a function of parameters (independent variables, factors, predictor variables) can be divided into global regression and smoothing (local regression). Neither of these methods is adequate, especially when the observations are a function of a time evolution variable and are probably highly correlated. The authors examine the use of the multivariate adaptive regression spline (MARS) methodology for this smoothing and characterization problem and the use of this methodology when there is serial correlation in the data so that lagged values of the observation can be used for predictor variables. The methodology is also useful when analyzing inputs to queues. The modeling of chemical warfare is considered as an example