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

Area of Science:

• Chemometrics
• Spectroscopy
• Data Analysis

Background:

• Partial Least Squares (PLS) is crucial in chemometrics for handling multicollinearity and high-dimensional data.
• Standard PLS models can struggle with highly complex datasets, limiting predictive performance.
• Local PLS addresses these limitations by focusing on relevant subsets of data.

Purpose of the Study:

• To provide a comprehensive tutorial on implementing local Partial Least Squares (PLS) methods.
• To enhance the accessibility and adoption of local PLS in chemometrics and spectroscopy.
• To offer practical guidance and ready-to-use code for researchers.

Main Methods:

• Local PLS methodology involves selecting nearest neighbors for each new sample.
• A PLS model is calibrated on this local subset for prediction.
• Implementation details and best practices are explained step-by-step.

Main Results:

• Local PLS ensures predictions are based on a more relevant local context.
• The approach is valuable for large and complex spectroscopic datasets.
• The tutorial provides practical examples and code in R and Julia.

Conclusions:

• Local PLS offers improved performance over standard PLS for complex datasets.
• This tutorial bridges the gap in accessible resources for local PLS implementation.
• The provided scripts facilitate the integration of local PLS into research workflows.