Asymptotically Optimal Regression Trees

Erik Mohlin ()

No 2018:12, Working Papers from Lund University, Department of Economics

Abstract: Regression trees are evaluated with respect to mean square error (MSE), mean integrated square error (MISE), and integrated squared error (ISE), as the size of the training sample goes to infinity. The asymptotically MSE- and MISE minimizing (locally adaptive) regression trees are characterized. Under an optimal tree, MSE is O(n^{-2/3}). The estimator is shown to be asymptotically normally distributed. An estimator for ISE is also proposed, which may be used as a complement to cross-validation in the pruning of trees.

Keywords: Piece-Wise Linear Regression; Partitioning Estimators; Non-Parametric Regression; Categorization; Partition; Prediction Trees; Decision Trees; Regression Trees; Regressogram; Mean Squared Error (search for similar items in EconPapers)
JEL-codes: C14 C38 (search for similar items in EconPapers)
Pages: 27 pages
Date: 2018-05-22
New Economics Papers: this item is included in nep-ecm
References: View references in EconPapers View complete reference list from CitEc
Citations:

Downloads: (external link)
https://lucris.lub.lu.se/ws/portalfiles/portal/194854013/WP18_12 Full text (application/pdf)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:hhs:lunewp:2018_012

Access Statistics for this paper

More papers in Working Papers from Lund University, Department of Economics School of Economics and Management, Box 7080, S-22007 Lund, Sweden. Contact information at EDIRC.
Bibliographic data for series maintained by Iker Arregui Alegria ().