A Near-infrared spectroscopy (NIRS) calibration for predicting nutrient quality in processed unconventional feedstuffs

The objective of this study was to develop NIRS calibrations to predict the quality of unconventional feedstuffs, focusing on crude fiber, crude protein, extract ether, and ash content. A total of 30 samples obtained from the fermentation process of citronella residues were used to create calibrated datasets. NIR spectra were measured using a BUCHI NIRFlex N-500 spectrometer, and a calibration model was developed with Unscrambler X 10.4 software, using Partial Component Regression (PCR). Cross-validation was utilised to optimise the model and prevent overfitting. Furthermore, Multiplicative Scatter Correction (MSC) and De-Trending (DT) techniques were employed to enhance the accuracy of the model. The results of the study indicated that the PCR model, utilising corrected spectra, yielded the most precise predictions. The model for ash content demonstrated a high degree of accuracy (R2 = 0.88; RPD = 3.06), while the predictions of crude protein were deemed adequate (R2 = 0.74; RPD = 2.03). However, predictions for crude fibre (R2 = 0.55; RPD = 1.54) were approximate, and those for extract ether (R2 = 0.41; RPD = 1.34) were less reliable. In conclusion, NIRS with proper preprocessing proves to be a reliable alternative to chemical tests for analyzing the quality of unconventional feedstuffs.