Non-destructive detection of multiple fatty acids in fuzzy cottonseeds based on near-infrared spectroscopy and chemometrics.

As an important oil resource, the fatty acids composition of cottonseed significantly influences its nutritional value and industrial application. This study develops a rapid and non-destructive detection method for nine kinds of fatty acids of fuzzy cottonseeds based on near-infrared spectroscopy (NIRS) coupled with chemometrics. To enhance the performance of the model, proposed the optimization strategy by integrating three preprocessing methods: derivative, Savitzky - Golay smoothing, and standard normal variate, and confirming parameters by grid searching. Moreover, employs SHapley Additive exPlanations (SHAP) analysis to improve the interpretability of the model. Results demonstrates that the R2 values of nine kinds of fatty acids (myristic acid, palmitic acid, palmitoleic acid, stearic acid, oleic acid, linoleic acid, α-Linolenic acid, arachidic acid, and total fatty acid) are 0.9454, 0.9050, 0.9524, 0.9617, 0.9224, 0.9520, 0.9043, 0.9393 and 0.9767. The values of relative performance deviation (RPD) and relative error rate (RER) are all exceeding 2.5 and 10. Screened wavelengths by SHAP can not only correspond to the molecular structures of the fatty acids, but also agree with the existing literature reports. The above results indicate that the models constructed can be applied to predict the content of nine kinds of fatty acids in fuzzy cottonseeds, providing technical support for the online detection of fatty acids in cottonseeds and their by-products. Furthermore, the integration of SHAP analysis enhances the interpretability of models, offering traceable support for model explanation.