Calculating the Dependency of Components of Observable Nonlinear Systems Using Artificial Neural Networks
Wolfram Rinke · RePEc: Research Papers in Economics · 2015
A method for computing dependency factors between independent and dependent components of an observable system is presented. The dependency factors are derived from the analysis of an artificial neural network, which models the system observed. The generic approach applies already proven methodologies to calculate the input-output sensitivity of multilayer feedforward artificial neural networks with differentiable activation functions. The algorithm and the mathematics to compute the dependency factors are presented. Different examples from agriculture and marketing illustrate how this method can be applied to explain an observable system.