The x-axis shows the number of parameters in the respective models. The y-axis shows the part of the output variance, which is NOT explained by the model. That is, the ratio between the prediction error variance and the output variance in percent.
Only that model that has the best fit for the given number of parameters is displayed. The fit is calculated using the Validation Data set.
Some text and color coding of bars relates some standard best choices:
Click on any of the bars using the (right) mouse button and the associated number of parameters will be inserted into the edit box. Alternately, type in any value you want. Further information about the best model with that number of parameters is displayed. If you are using this facility together with the GUI, press Insert to insert the model into the ident Model board. If you are using it by calling selstruc from the command line, press Select, and a variable nn containing the chosen model structure is created in the workspace. You can make several selections.
Theory suggests that Best Fit should be chosen if the Validation Data set is different from the Working Data set. Either AIC or MDL should be chosen if the Working and Validation data sets coincide. It may be wise to make a more subjective trade-off between fit and model order.
(file selordax.htm)