RESEARCH ARTICLE Local partial least squares based on global PLS scores Guanghui Shen 1,2 | Matthieu Lesnoff 3,4 | Vincent Baeten 2 | Pierre Dardenne 2 | Fabrice Davrieux 5,6 | Hernan Ceballos 7 | John Belalcazar 7 | Dominique Dufour 6,7,8,9 | Zengling Yang 1 | Lujia Han 1 | Juan Antonio Fernández Pierna 2 1 College of Engineering, China Agricultural University, Beijing, China 2 Food and Feed Quality Unit, Valorisation of Agricultural Products Department, Walloon Agricultural Research Center, Gembloux, Belgium 3 CIRAD, UMR SELMET, Montpellier, France 4 SELMET, Université de Montpellier, CIRAD, INRA, Montpellier Sup Agro, Montpellier, France 5 UMR Qualisud, CIRAD, Saint‐Denis, France 6 Qualisud, Université de Montpellier, CIRAD, Montpellier SupAgro, Université d'Avignon, Université de La Réunion, Montpellier, France 7 Cassava Program, International Center for Tropical Agriculture (CIAT), Cali, Colombia 8 UMR Qualisud, CIRAD, Cali, Colombia 9 UMR Qualisud, CIRAD, Montpellier, France Correspondence Juan Antonio Fernández Pierna, Food and Feed Quality Unit, Valorisation of Agricultural Products Department, Walloon Agricultural Research Center, Gembloux, Belgium. Email: foodfeedquality@cra.wallonie.be; j. fernandez@cra.wallonie.be Funding information China Scholarship Council; HarvestPlus; CIRAD; CIAT Abstract A local‐based method for near‐infrared spectroscopy predictions, the local partial least squares regression on global PLS scores (LPLS‐S), is proposed in this work and compared with the usual local PLS (LPLS) regression approach. LPLS‐S is based on the idea of replacing the original spectra with a global PLS score matrix before using the usual LPLS. This is done with the aim of increas- ing the speed of the calculations, which can be an important parameter for online applications in particular, especially when implemented on large databases. In this study, the performance of the two local approaches was compared in terms of efficiency and speed. It could be concluded that the root‐mean‐square error of prediction of LPLS and LPLS‐S were 1.1962 and 1.1602, respectively, but the calculation speed for LPLS‐S was more than 20 times faster than for the LPLS algorithm. KEYWORDS local calibration, near‐infrared spectroscopy, partial least squares, running speed 1 | INTRODUCTION Near‐infrared spectroscopy (NIRS), well known as a fast and nondestructive analytical method, is widely used in combination with partial least squares (PLS) for quality control and shows great potential for application in many fields, such as food (including fruits, 1-3 dairy products, 4 tea, 5 meat, 6 fish, 7 and grain 8,9 ), pharmaceuticals, 10 biomass, 11 Received: 24 October 2018 Revised: 19 December 2018 Accepted: 16 January 2019 DOI: 10.1002/cem.3117 Journal of Chemometrics. 2019;e3117. https://doi.org/10.1002/cem.3117 © 2019 John Wiley & Sons, Ltd. wileyonlinelibrary.com/journal/cem 1 of 12