The valorization of agricultural residues as alternative feed resources is increasingly critical for enhancing livestock sustainability. This study investigates the potential of sequential fermentation to improve the nutritional quality of citronella (Cymbopogon nardus L.) residues and evaluates the use of near-infrared spectroscopy (NIRS) combined with machine learning (ML) models for rapid feed quality assessment. Citronella residues were subjected to sequential fungal and lactic acid bacterial fermentation, and their feed quality attributes, including moisture, crude protein, crude fiber, ether extract, and ash content, were measured using standard laboratory methods. NIR spectra were acquired from 1000 to 2500 nm and analyzed using partial least squares regression (PLSR), ridge regression, adaptive boosting (AdaBoost), and support vector machine regression (SVMR). Principal component analysis (PCA) revealed a high degree of spectral homogeneity with sufficient underlying variability to enable robust modeling. Among the models evaluated, AdaBoost and SVMR consistently outperformed linear models, achieving high coefficients of determination (R² ≥ 0.99) and low root mean square errors (RMSE). Particularly, SVMR and AdaBoost achieved high predictive accuracy for moisture, crude protein, and ether extract content, with residual predictive deviation (RPD) values far exceeding standard thresholds. The integration of sequential fermentation, NIRS, and advanced ML algorithms presents a rapid, non-destructive, and sustainable approach to upgrading and monitoring alternative fibrous feed sources, supporting broader initiatives in circular bioeconomy and sustainable animal production.