A step-by-step tutorial to local partial least squares analyses for near-infrared spectroscopic data analysis

Partial least squares (PLS) methods are key algorithms in chemometrics, particularly in near-infrared spectroscopy. Their ability to handle multicollinearity and high-dimensional data makes them an effective tool for regression and discrimination tasks. Nevertheless, PLS performance can be limited when dealing with highly complex datasets. To address this, local PLS analyses were introduced. Instead of calibrating a model on the entire dataset and ignoring local variations, local PLS selects a subset of nearest neighbours for each new sample to predict and calibrate a PLS model on this subset. This ensures that each sample is predicted using a model trained on a more relevant local context. Such an approach is easy to implement and especially valuable with the growing availability of large spectroscopic datasets. Despite its advantages, local PLS remains underutilized, perhaps due to a lack of accessible resources. This tutorial aims to bridge that gap by providing a step-by-step framework for implementing local PLS, combining theoretical insights with hands-on applications. Readers will gain practical knowledge through detailed explanations, examples, and best practices for avoiding common mistakes. Ready-to-use scripts written in R and Julia languages are provided to facilitate adoption and integration into research workflows.