Open Collections

Small sample improvement over Bayes prediction under model uncertainty

University of British Columbia

Existing criteria for evaluating the adequacy of a predictive model are model-based (e.g. AIC, BIC, MSPE) or empirical (e.g. PRESS and other cross-validation type criteria). We introduce a new class of "mongrel" criteria for on-line prediction that evaluates candidate predictors based on both model information and past empirical performance. Simulation results showed that the mongrel procedure produced more accurate predictions than the standard Bayes procedure for small sample sizes. This improvement was observed over a wide range of data-generators for the problem of variable selection in normal linear models.

Thesis/Dissertation

application/pdf

2009-07-23

For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use.

http://hdl.handle.net/2429/11210

Statistics

University of British Columbia

UBCV

DSpace