Unformatted text preview: Akaike Information Criterion AIC.
AIC = −2l (θ; Y ) + 2p
where p is the number of parameters in the statistical
Select model with minimum AIC . UNM For our example,
AIC0 = 2(68.3868) + 2(1) = 138.7736
AIC1 = 2(67.0230) + 2(2) = 138.046
So the preferred model is θ1 = θ2 but barely.
AIC rewards goodness of ﬁt or models with large likelihood
Includes penalty that increases with number of parameters.
Attempts to avoid overﬁtting.
Different penalties provide different criteria,
View Full Document
- Fall '13
- Trigraph, Maximum likelihood, Errors and residuals in statistics, Akaike information criterion