A hybrid preprocessing strategy-based NIR spectroscopy method for rapid determination of nutritional components in Cyperus esculentus with SHAP interpretation.

Cyperus esculentus, a plant rich in unsaturated fatty acids, is characterized by high yield, nutritional value, and drought resistance. Rapid detection of its nutritional components is of great significance for ensuring food security and ecological restoration. In this study, near-infrared (NIR) spectroscopy, in combination with chemometrics, was employed to analyze the characteristics and the contents for crude fat (CF), total sugar (TS), and crude protein (CP) of Cyperus esculentus. To address spectral variability and sample heterogeneity, a hybrid strategy combining Savitzky-Golay smoothing, and standard normal variate was proposed. Grid search was used for parameter optimization, with SHapley Additive Explanations (SHAP) analysis applied to identify the most interpretable variables. The results revealed marked differences in the nutritional profiles of Cyperus esculentus across regions, indicating the potential to develop region-specific products based on these characteristics. For test set, the prediction models achieved high coefficient of determination (R2) of 0.9847 (CF), 0.9096 (TS), 0.9876 (CP). These findings demonstrate that the proposed method enables rapid and accurate prediction of nutritional components in Cyperus esculentus. Notably, the chemical variables screened by SHAP exhibit a high degree of alignment with nutritional profiles, significantly enhancing model interpretability. The proposed method enables rapid and accurate prediction of these components in Cyperus esculentus with high interpretability robustness, thereby establishing an innovative technical framework for quality assessment.