Ridge partial correlation screening for ultrahigh-dimensional data

Variable selection in ultrahigh-dimensional linear regression ischallenging due to its high computational cost. Therefore, ascreening step is usually conducted before variable selection tosignificantly reduce the dimension. Here we propose a novel andsimple screening method based on ordering the absolute sample ridgepartial correlations. The proposed method takes into account notonly the ridge regularized estimates of the regression coefficientsbut also the ridge regularized partial variances of the predictorvariables providing sure screening property without strongassumptions on the marginal correlations. Simulation study and areal data analysis show that the proposed method has a competitiveperformance compared with the existing screening procedures. Apublicly available software implementing the proposed screeningaccompanies the article.

@article{wang2025_2504.19393,
  title={ Ridge partial correlation screening for ultrahigh-dimensional data },
  author={ Run Wang and An Nguyen and Somak Dutta and Vivekananda Roy },
  journal={arXiv preprint arXiv:2504.19393},
  year={ 2025 }
}