Over-Fitting
From Open Risk Manual
Definition
Over-fitting is a potential pathology of the Model Development process, whereby a model that is too flexible (e.g., too many parameters or degrees of freedom) is calibrated to reproduce very closely the available data.
Mitigation Strategies
For statistical models a good strategy is to monitor the Akaike Information Criterion
Issues and Challenges
- For many classes of models there are no theoretically based criteria to help avoid over-fitting
