A State-Based Sensitivity Analysis for Distinguishing the Global Importance of Predictor Variables in Artificial Neural Networks
Outcomes of a multilayer perceptron for a representative simulation (Model 37; see text) within Case Study. (a) Mean square error associated with training and cross-validation data subsets during network training. (b and c) Modeled values as a function of actual values in (b) training and cross-validation data ( and 22, resp.). (c) Test data () subsets. The dashed line represents a 1 : 1 relationship.