Splitting variable selection for multivariate regression trees
We show that the usual exhaustive search principle adapted for multivariate regression trees has selection bias toward variables with more split points. A selection scheme is proposed to control bias by utilizing hierarchical loglinear model for three-way contingency table of residuals.
Year of publication: |
2007
|
---|---|
Authors: | Hsiao, Wei-Cheng ; Shih, Yu-Shan |
Published in: |
Statistics & Probability Letters. - Elsevier, ISSN 0167-7152. - Vol. 77.2007, 3, p. 265-271
|
Publisher: |
Elsevier |
Subject: | Bias Loglinear model Residual |
Saved in:
Saved in favorites
Similar items by person
-
Interval Estimation for a First-Order Positive Autoregressive Process
Hsiao, Wei-Cheng, (2018)
-
Split variable selection for tree modeling on rank data
Kung, Yi-Hung, (2012)
-
Selecting the best splits for classification trees with categorical variables
Shih, Yu-Shan, (2001)
- More ...