Difference between revisions of "Flat Maximum Effect"
From Open Risk Manual
Wiki admin (talk | contribs) |
Wiki admin (talk | contribs) |
||
| Line 2: | Line 2: | ||
The '''Flat Maximum Effect''' is an informal term capturing a potential pathology (weakness) of statistical [[Risk Model | models]] where models with different coefficients or even belonging to different model classes can have similar performance on prediction or classification. | The '''Flat Maximum Effect''' is an informal term capturing a potential pathology (weakness) of statistical [[Risk Model | models]] where models with different coefficients or even belonging to different model classes can have similar performance on prediction or classification. | ||
| − | The term originates from the shape of the error function or the [[Likelihood Function] when it exhibits reduced sensitivity to model parameters near its maximum | + | The term originates from the shape of the error function or the [[Likelihood Function]] when it exhibits reduced sensitivity to model parameters near its maximum |
== Impact == | == Impact == | ||
Latest revision as of 14:52, 30 September 2021
Definition
The Flat Maximum Effect is an informal term capturing a potential pathology (weakness) of statistical models where models with different coefficients or even belonging to different model classes can have similar performance on prediction or classification.
The term originates from the shape of the error function or the Likelihood Function when it exhibits reduced sensitivity to model parameters near its maximum
Impact
- Difficulty selecting the most appropriate model
- Difficulty to interpret (explain) outcomes in terms of model parameters
