Goodness of Fit Calculations

Goodness of Fit metrics are calculated for learning points, for verification points, and by cross-validating learning points.

Learning Points

Goodness of Fit for learning points (the DOE points and refinement points used to generate the response surface) measures the quality of the response surface interpolation. To calculate Goodness of Fit for learning points, the error between the predicted and observed values is calculated for each point.

Verification Points

Goodness of Fit for verification points measures the quality of the response surface prediction. To calculate Goodness of Fit for verification points (which are not used in the building of the response surface), the verification points are placed in locations that maximize the distance to the learning points. After the verification points are calculated with a design point update, the differences between verification point values and predicted values are calculated.

For more information, see Verification Points.

Learning Points with Cross-Validation

Goodness of Fit obtained via the cross-validation technique measures the stability and reliability of the response surface. To calculate Goodness of Fit for cross-validation learning points, the cross-validation computes the error between the observed value and the predicted value for each point. However, the predicted value is evaluated on a model built with all DOE points except the associated observed value. For example, the first response surface is built without the first design point, the second response surface is built without the second design point, and so on.

Assume the following notation:

The standard GOF on learning points compares to when the GOF based on the cross-validation compares to .

For more information, see the Genetic Aggregation theory section.


Release 18.2 - © ANSYS, Inc. All rights reserved.