A Tutorial on Chemometrics for Low Field NMR Relaxation Data Part 2: Cluster, Regression and Classification Analysis

Authors:
Preprint
Publisher website
Google Scholar
Abstract
LF-NMR signals, typically acquired as multi-exponential decays, contain rich information about molecular mobility, compartmentalization and sample composition (i.e., T₂ or T₁). However, their full potential is often underutilized due to a lack of chemometric integration. This tutorial begins with Exploratory data analysis using principal component analysis and hierarchical cluster analysis, emphasizing preprocessing strategies such as mean-centering and normalization to reveal latent structures. A case study on temperature-dependent egg measurements illustrates how preprocessing choices affect physicochemical interpretation. Next, regression methods, univariate, multiple linear regression, principal component regression, and partial least squares (PLS), are detailed with practical examples on pharmaceutical and food adulteration datasets, highlighting the importance of cross-validation, figures of merit (RMSE, R², LOD, LOQ, bias), and model robustness. Finally, classification techniques including K-nearest neighbors, PLS Discriminant Analysis (PLS-DA), Soft Independent Modeling of Class Analogy (SIMCA), and one-class data driven SIMCA are demonstrated on the same egg dataset. The tutorial concludes with emerging trends (i.e., non-linear models, SHapley Additive explanations interpretability, data fusion, class balancing) and best practices for building causal, interpretable, and generalizable LF-NMR chemometric models