Skip to Main content Skip to Navigation
Conference papers

Décorrélation adaptative pour la prédiction en grande dimension

Abstract : In large-scale signicance analysis, ignoring dependence or not is a core issue, leading to many recent results about the impact of decorrelating the pointwise test statistics. Yet, for the estimation of a prediction model, decorrelating large proles of predicting variables is not as clearly questioned, although many comparative studies have reported the superiority of so-called naive methods, ignoring dependence. Under the usual Gaussian mixture model assumption of Linear Discriminant Analysis, we show that, for a given dependence structure, the classication performance of methods ignoring or not dependence may be markedly dierent, according to the pattern of the association signal between the predicting variables and the response. In order to minimize the largest probability of misclassication, we propose a method handling adaptively the dependence. A simulation study shows that the performance of the present method is at least as good as the best of methods ignoring dependence or based on a complete decorrelation of the predicting variables. 1
Document type :
Conference papers
Complete list of metadata

Cited literature [4 references]  Display  Hide  Download
Contributor : Catherine Cliquet Connect in order to contact the contributor
Submitted on : Wednesday, November 13, 2019 - 3:13:10 PM
Last modification on : Friday, May 20, 2022 - 9:04:50 AM


Files produced by the author(s)


Distributed under a Creative Commons Attribution - NonCommercial - NoDerivatives 4.0 International License


  • HAL Id : hal-02361735, version 1


Florian Hébert, Mathieu Emily, David Causeur. Décorrélation adaptative pour la prédiction en grande dimension. 51es Journées de Statistique 2019, Société Française de Statistique, Jun 2019, Nancy, France. ⟨hal-02361735⟩



Record views


Files downloads