In addition to assessing accuracy for each model, rele- vant features were examined to determine the important variables in the model. For the ANN model, sensitivity analysis was used. In ANN, sensitivity analysis rates pre- dictor variables according to the deterioration in modeling performance that occurs if that variable is no longer avail- able to the model. The basic measure of sensitivity of a pre- dictor variable is calculated as the ratio of the error of the model without the inclusion of the variable to the error of the model that included the variable. The more sensitive the network is to a particular input, the greater the deteri- oration one can expect, and therefore the greater the ratio. The shortcoming of this approach is that it assumes the independent contribution of variables to the outcome of the model, which may not hold true in situations with inter- dependent variables that are important only if included as a set. For the decision tree model, variable importance mea- sures were used to judge the relative importance of each predictor variable. Variable importance ranking uses surro- gate splitting to produce a scale (a relative importance mea- sure) for each predictor variable included in the analysis. The computational details regarding these measures can be found in Breiman et al. (1984). The variable importance results are shown in Table 3.