2.8.   Too Much Information Deteriorates Calibration

In chapter 5, a feature extraction is presented which generates many variables. The increase of information puts new challenges to multivariate calibration methods and especially to the data analysis by neural networks. Section 2.8.1 describes the fundamental relationships between the model complexity and the generalization and prediction abilities of the models, which are directly influenced by the number of independent variables utilized by the models. In section 2.8.2, the aspects, which influence the complexity of neural networks, are looked at. Then the different approaches for the optimization of the model complexity found in literature are discussed in the following sections.