One-sided cross-validation for nonsmooth regression functions

Olga Y. Savchuk, Jeffrey D. Hart, Simon P. Sheather

Research output: Contribution to journalArticlepeer-review

2 Scopus citations


The one-sided cross-validation (OSCV) method is shown to be robust to lack of smoothness in the regression function. Two corrections for the case where the regression function has a discontinuous first derivative are proposed. Simulation results suggest that proposed modifications of the OSCV method are efficient for regression functions whose first derivative is discontinuous at more than two points. The OSCV method and its modification outperform the cross-validation method and the Ruppert-Sheather-Wand plug-in method in a data example involving a function that, potentially, has one discontinuity in its derivative.

Original languageEnglish
Pages (from-to)889-904
Number of pages16
JournalJournal of Nonparametric Statistics
Issue number4
StatePublished - Dec 2013

Bibliographical note

Funding Information:
The authors thank two anonymous referees whose comments led to the substantial improvement of the article. The work of Professor Hart was supported by NSF grant DMS 1106694.


  • average squared error
  • bandwidth selection
  • cross-validation
  • local linear estimator
  • mean average squared error

ASJC Scopus subject areas

  • Statistics and Probability
  • Statistics, Probability and Uncertainty


Dive into the research topics of 'One-sided cross-validation for nonsmooth regression functions'. Together they form a unique fingerprint.

Cite this