$L_2$Boosting and model selection
碩士 === 國立中正大學 === 統計科學所 === 95 === In this thesis, we first study $L_2$Boosting with model selection criteria such as AIC_c, BIC, and gMDL, then review Buhlmann and Yu''s (2005) work. In Buhlmann and Yu''s (2005) work, when using model selection criterion to estimate the stopping...
| Main Authors: | , |
|---|---|
| Other Authors: | |
| Format: | Others |
| Language: | en_US |
| Online Access: | http://ndltd.ncl.edu.tw/handle/99374430668180483548 |
| Summary: | 碩士 === 國立中正大學 === 統計科學所 === 95 === In this thesis, we first study $L_2$Boosting with
model selection criteria such as AIC_c, BIC, and gMDL, then
review Buhlmann and Yu''s (2005) work. In Buhlmann and Yu''s (2005)
work, when using model selection criterion to estimate the stopping
iteration for $L_2$Boosting approach, it is necessary to compute all
boosting iterations under consideration for the training data.
Hence, we propose a new CP-$L_2$Boosting which is an approach based
on detecting the change point of the model selection criteria to
seek the earlier stopping iterations of the training data in
$L_2$Boosting procedure. We also extend this new method to
classification problems and compare them with LogitBoost by using
apparent classification rates and area under ROC curve (AUC)
criteria. The simulation studies and a real data example illustrate
that the method proposed in this thesis can make a substantial
computational saving.
|
|---|