Analyzing mathematical models with inductive learning networks

After building and validating a mathematical model, the decision maker frequently solves (often many times) a slightly different version of the model. That is, by changing various input parameters and re-running different model instances, he tries to develop insight(s) into the workings and tradeoffs of the complex system represented by the model. However, very little research has been devoted to helping the decision maker in this important model analysis endeavor. This paper investigates the application of two inductive learning technologies, backpropagation neural networks and the group method of data handling, to the analysis of multiple instances of a mathematical model. Specifically, these two techniques are compared in the analysis tasks of identifying key factors and determining key relations between uncertain and/or unknown model parameters and the associated objective function values.