Prediction of Physicochemical Properties of Date Fruits from LF-NMR Relaxation: Regressions, Multivariate Analysis, and Machine Learning

Simple prediction of food composition and related properties could reduce reliance on complicated measurement methods that use costly instruments and chemicals. In this study, low-field nuclear magnetic resonance (LF-NMR) relaxation measurements (T1 and T2) were used. Nine batches of date fruits (DFs) differing in quality were analyzed using relaxation descriptors extracted to characterize proton environments associated with different mobility states, with transverse (T₂) relaxation reflecting contributions from rigid, semirigid, and mobile proton populations. Descriptors associated with rigid, semirigid, and mobile proton populations were obtained using transverse (T2) relaxation. The relationships between these parameters and the physicochemical characteristics of the samples of date fruit were analyzed. LF-NMR T1 and T2 parameters showed linear correlations with studied physicochemical properties, including moisture, crude fiber, and pectin. This indicates the potential of LF-NMR for predicting physicochemical properties of date fruits. Multivariate analysis was used to examine clustering trends among date fruits based on LF-NMR relaxation behaviors. The decision tree model achieved high predictive performance for moisture content (R2 = 0.990) using T₂ relaxation parameters. Artificial neural networks considering T2 relaxation also showed strong internal predictive performance for sucrose (R2 = 0.998) and fiber (R2 = 1.000). These results demonstrate that integrating LF-NMR relaxation with multivariate and machine learning approaches provides an effective framework for predicting key physicochemical attributes of date fruits.