Scalable bayesian bi-level variable selection in generalized linear models
Résumé
Motivated by a real-world application in cardiology, this paper introduces a novel algorithm designed for Bayesian bi-level variable selection in generalized linear models. The methodology is specifically tailored to address the challenges posed by large datasets, characterized by both a substantial number of individuals and predictors. Our approach relies on the waste-free Sequential Monte Carlo (SMC) methodology developed by [3]. Notably, we address the intricate constraints inherent in bi-level variable selection, where the selection of an individual predictor is contingent upon the prior selection of its group. To tackle this challenge, we present a new proposal mechanism that effectively navigates these constraints, ensuring a robust and efficient variable selection process. Additionally, our algorithm incorporates the Approximate Laplace Approximation (ALA) approach proposed by [11], enhancing the computational efficiency of our method. We demonstrate in our numerical study that the algorithm consistently offers reliable performance on large datasets within a few minutes, on both simulated data and real-world data related to the aforementioned cardiology application.
Domaines
Statistiques [math.ST]Origine | Publication financée par une institution |
---|