Strategy for constructing calibration sets based on a derivative spectra information space consensus

• An improved Kennard-Stone calibration set construction strategy is proposed for the multivariate regression model.
• The core idea of strategy is to make full use of derivative spectra information when constructing the calibration set.
• A spectra estimator based on singular perturbation technique is introduced and used to construct derivative spectra space.
• The proposed strategy can give a more reasonable result of sample sets partitioning for the multivariate regression model.

Constructing an excellent calibration set is crucial to ensuring accurate multivariate calibration of spectra data. The purpose of this paper is to present an improved Kennard–Stone (KS) calibration set construction strategy based on different derivative spectra information spaces, termed Consensus Kennard–Stone (CKS). The core idea is to make full use of different derivative spectra information spaces when constructing the calibration set using a consensus selection method as well as to improve the prediction performance of the multivariate regression model. The experimental results from two public spectra datasets indicate that the proposed CKS strategy can use a more appropriate subset of samples for constructing the calibration set in the multivariate regression model and has superior predictive performance compared with the existing classic sample-selection KS strategies.

Figure optionsDownload as PowerPoint slide