De-noising boosting methods for variable selection and estimation subject to error-prone variables

• Published in: Statistics and computing Vol. 33; no. 2
• Main Author: Chen, Li-Pang
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.04.2023; Springer Nature B.V
• Subjects: Algorithms; Artificial Intelligence; Computer Science; Error analysis; Error correction; Original Paper; Probability and Statistics in Computer Science; Statistical methods; Statistical Theory and Methods; Statistics and Computing/Statistics Programs; SIMEX; Regression calibration; Error correction; Statistical learning; Mismeasurement; Generalized linear models
• ISSN: 0960-3174; 1573-1375
• DOI: 10.1007/s11222-023-10209-3

Boosting is one of the most powerful statistical learning methods that combines multiple weak learners into a strong learner. The main idea of boosting is to sequentially apply the algorithm to enhance its performance. Recently, boosting methods have been implemented to handle variable selection. However, little work has been available to deal with complex data such as measurement error in covariates. In this paper, we adopt the boosting method to do variable selection, especially in the presence of measurement error. We develop two different approximated correction approaches to deal with different types of responses, and meanwhile, eliminate measurement error effects. In addition, the proposed algorithms are easy to implement and are able to derive precise estimators. Throughout numerical studies under various settings, the proposed method outperforms other competitive approaches.