Wolfram Library Archive

Courseware Demos MathSource Technical Notes
All Collections Articles Books Conference Proceedings
Title Downloads

Selecting and Fitting Nonlinear Models to Time Series from Dynamic Systems--Handling the Overfitting Problem with the Neural Networks Package

Jonas Sj÷berg
Organization: Chalmers University of Technology

International Mathematica User Conference 2008
Conference location

Champaign, IL

When nonlinear models are fitted to data it is important to make sure that the data is not overfitted. The reason for this is that many possible model structures, all with many parameters, are fitted to the training data, then there is a high risk that the model picks up the specific disturbances in the training data rather than the general aspects in the data generating process.

The talk illustrates the problem and gives some general possibilities how this problem can be handled. Especially, the methods supported by the Neural Network package together with general support in Mathematica is illustrated.

The talk partly builds on excerpts from the course WEG 330.

*Wolfram Technology

Downloads Download Wolfram CDF Player

SelectingAndFittingNonLinearModels_Abstract.nb (250.9 KB) - Mathematica Notebook
SelectingAndFittingNonLinearModels_Presentation.nb (963.9 KB) - Mathematica Notebook