Abstract
Artificial Neural Network (ANNs) are largely used to replace computational expensive models. However, these surrogate models can introduce additional uncertainty and variability on the output of interest. In particular, when it is used to estimate the sensitivity of a model, the result obtained might give a false confidence, when the analyst is not aware of the uncertainty introduced. Hence, it is of fundamental importance to first check the validity of the ANN, and then quantify the uncertainties associated with the point estimates of the model. In this paper, an ANN is constructed based on selected data representative of the input/output non-linear relationship of an underlying waste management model (SIXEP). Once constructed, the ANN is used for performing sensitivity analysis in reasonable computational time. Finally, the bootstrapped technique is adopted to quantify the uncertainty introduced by the surrogate model in terms of bias corrected confidence intervals.