EUVIMEDEUVIMED entdecken
Crossref

An iterative matrix uncertainty selector for high-dimensional generalized linear models with measurement errors

Betrand Fesuh Nono iD, Georges Nguefack-Tsague, Martin Kegnenlezom, Eugène-Patrice N Nguéma

Statistical Methods in Medical Research · 2025

DOI: 10.1177/09622802251316963 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Statistical Methods in Medical Research, 2025. Verfasst von Betrand Fesuh Nono und 3 weiteren Autor:innen. Beteiligte Einrichtung: National Advanced School of Engineering, University of Yaoundé I, Cameroon. Hauptthemen laut MEDLINE: Algorithms. Die Arbeit nennt 32 Literaturangaben.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Measurement error is a prevalent issue in high-dimensional generalized linear regression that existing regularization techniques may inadequately address. Most require estimating error distributions, which can be computationally prohibitive or unrealistic. We introduce an error distribution-free approach for variable selection called the Iterative Matrix Uncertainty Selector (IMUS). IMUS employs the matrix uncertainty selector framework for linear models, which is known for its selection consistency properties. It features an efficient iterative algorithm easily implemented for any generalized linear model within the exponential family. Empirically, we demonstrate that IMUS performs well in simulations and on three microarray gene expression datasets, achieving effective covariate selection with smoother convergence and clearer elbow criteria compared to other error distribution free methods. Notably, simulation studies in logistic and Poisson regression showed that IMUS exhibited smoother convergence and clearer elbow criteria, performing comparably to the Generalized Matrix Uncertainty Selector (GMUS) and Generalized Matrix Uncertainty Lasso (GMUL) in covariate selection. In many scenarios, IMUS had smaller estimation errors than GMUL and GMUS, measured by both the 1- and 2-norms. In applications to three microarray datasets with noisy measurements, GMUS faced convergence issues, while GMUL converged but lacked well-defined elbows for two datasets. In contrast, IMUS converged with well-defined elbows for all datasets, providing a potentially effective solution for high dimensional regression problems involving measurement errors.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

AlgorithmsHumansComputer SimulationGene Expression ProfilingLinear ModelsLogistic ModelsPoisson DistributionUncertainty

Stichwörter der Autor:innen

Generalized linear modelgene expressionhigh-dimensional dataiterative re-weighted least squaresmatrix uncertainty selectormeasurement error

Autor:innen und Institutionen

  • National Advanced School of Engineering, University of Yaoundé I, Cameroon.
  • Faculty of Medicine and Biomedical Sciences, University of Yaoundé I, Cameroon.
  • Department of Mathematics, University of Yaoundé I, Cameroon.

ORCID-Profile: Betrand Fesuh Nono

Zitieren

APA

Fesuh Nono, B., Nguefack-Tsague, G., Kegnenlezom, M., & Nguéma, E. N. (2025). An iterative matrix uncertainty selector for high-dimensional generalized linear models with measurement errors. Statistical Methods in Medical Research. https://doi.org/10.1177/09622802251316963

Vancouver

Fesuh Nono B, Nguefack-Tsague G, Kegnenlezom M, Nguéma EN. An iterative matrix uncertainty selector for high-dimensional generalized linear models with measurement errors. Statistical Methods in Medical Research. 2025. doi:10.1177/09622802251316963

BibTeX
@article{FesuhNono2025,
  author = {Fesuh Nono, Betrand and Nguefack-Tsague, Georges and Kegnenlezom, Martin and Nguéma, Eugène-Patrice N},
  title = {An iterative matrix uncertainty selector for high-dimensional generalized linear models with measurement errors},
  journal = {Statistical Methods in Medical Research},
  year = {2025},
  doi = {10.1177/09622802251316963}
}
Kostenlos starten →