Algorithms for Robust Predictor Filtering and Evaluation of Their Stability

Abstract

The paper is devoted to the development of an algorithm for reliable predictor filtering based on Henze-Zirkler statistics, development of an iterative procedure based on Lass regularization. The stability of the algorithms based on modelled and real examples is studied. The description and investigation of existing robust filtering algorithms are given. In the process, two algorithms have been implemented for the study. The second procedure is an improvement of the first algorithm which screens out highly correlated predictors. A comparative analysis with existing filtering algorithms was carried out, and the stability of Henze-Zirkler robust filtering algorithm and the iterative procedure based on Lasso regularization was investigated. As a result of the work, conclusions were drawn about the effectiveness of the Henze-Zirkler robust filtering algorithm and the iterative procedure based on Lasso regularization. In addition, shortcomings in the stability of the algorithms when dealing with categorical data were identified.