Interval selection: A case‐study‐based approach

Abstract

Variable selection plays a fundamental role in the analysis of data containing several variables which are redundant or irrelevant to the problem of interest. The ability to identify and discard these variables would make it possible to improve predictive performances and data interpretation, thus reducing costs and computational time. Although many methods have been proposed for feature selection, in some fields there is more interest in selecting groups of variables because of the continuous nature and covariance of adjacent data. This is the case for near‐infrared spectroscopy, where several methods, mainly based on partial least squares regression, have been proposed to deal with interval selection. In this article, we consider some of these methods and propose an additional solution based on a variable clustering procedure (Cov/VSURF), Lasso regression and permutation tests. We compare their performances on four different public datasets and discuss the impact of interval selection on the predictive performances of the considered models.

Design choice and machine learning model performances

“…Additionally, the considered nonparametric ranking procedure also proved its efficacy from the point of view of the reliability of results as it has been extensively validated by means of simulation studies 53 and practical applications considering both observational and experimental data in several fields, 54 including medicine, new product development and marketing studies. Recently it has also been included as a core component in a methodology that performs variable selection in near‐infrared spectroscopy using ML models 55 …”

“…Recently it has also been included as a core component in a methodology that performs variable selection in near-infrared spectroscopy using ML models. 55 In this paper the permutation tests are applied using the difference in means as test statistics and assuming independent or paired data, depending on the specific situation. Considering 𝐺 𝑖 and 𝐺 𝑗 with 𝑖, 𝑗 = 1, … , 𝐶, 𝑖 ≠ 𝑗 two different groups of data to be compared (e.g., the different experimental designs), the permutation testing framework is employed to test the directional alternative hypothesis RMSE 𝐺 𝑖 > RMSE 𝐺 𝑗 , where RMSE is the prediction error calculated on the test data, i.e.…”