NIRS-based rapid analysis of crude fiber and crude protein in fermented citronella residues as non-conventional feedstuffs

Abstract Crude protein (CP) and crude fiber (CF) are important parameters of feed quality. Near-infrared spectroscopy (NIRS) has emerged as a valuable tool for evaluating the nutritional quality of feedstuffs. This study aimed to quickly assess the nutrient content of citronella residues as non-traditional fibrous feedstuffs using NIRS data. A total of 30 samples, subjected to various fermentation treatments, were used in calibrated datasets. Near-infrared spectra were obtained using the Thermo Nicolet Antaris II spectrometer. Calibration models were created using The Unscrambler X 10.4 software, employing the Partial Component Regression (PCR) and Partial Least Square Regression (PLSR) methods. 10-fold cross-validation was performed to validate the models. The first derivative with Savitzky-Golay smoothing (SGs) was utilized to enhance model accuracy and compare to the reliability of raw spectra in predicting models. The CP and CF contents of fermented citronella residues are predicted to be moderate and good by NIRS, with a maximum R 2 value of 0.54 and 0.85, respectively. The PLSR model demonstrated superior predictive capabilities compared to the PCR model. The PLSR model for predicting CF content showed adequate predictive stability and can forecast reliably, while the PLSR model for predicting CP content was suitable for screening and needed improvement.