Bypassing NIR pre-processing optimization with multiblock pre-processing ensemble approaches

Abstract

Pre-processing near-infrared spectral data is a major part of near-infrared data modelling. A wide range of pre-processings are available to deal with both the additive and the multiplicative effects. However, practitioners have majorly focused on the selection of the best pre-processing technique or their combination. Data pre-processed with different pre-processings carry complementary information; hence, a natural solution to avoid pre-processing selection and to learn complementary information is the ensemble modelling. Recently, multiblock data fusion modelling-inspired ensemble techniques have gained momentum and several innovative approaches have been proposed for modelling near-infrared data. This article provides a state of the art of the new multiblock modelling-inspired pre-processing ensemble techniques. Their novelties and pitfalls are also discussed.