Near-infrared spectroscopy for predicting chemical composition and classifying local chicken meat

Near-infrared (NIR) spectroscopy was tested to predict meat quality and classify chicken parts, sex, and diet in a local chicken breed. Sixty chicks (both sexes) were reared under identical conditions until 120 days, then assigned to a low-lipidic (LL, 3.85% EE, ether extract) and a high-lipidic diet (HL, 9.49% EE). At 150 days, breast and thigh samples were collected, and physical traits -pH and color- and proximate composition -moisture, crude protein (CP), ether extract (EE), and ash- were determined using wet chemistry methods. Samples were scanned intact, ground, and freeze-dried with a benchtop NIR spectrometer (1100-2500 nm), and prediction models were developed on both fresh and dry matter bases. External validation (70/30% split) was applied for pH, yellowness, EE, and ash, whereas cross-validation was performed separately for breast and thigh for redness, lightness, moisture, and CP. Model performance was evaluated using the coefficient of determination in validation (R²_VAL) and cross-validation (R²_CV), and the ratio of performance to deviation (RPD). Physical traits and ash showed poor predictability (R²_VAL and R² _CV < 0.68; RPD < 1.74), while chemical traits, particularly EE in freeze-dried samples, achieved excellent prediction (R²_VAL > 0.95; RPD > 3). Partial Least Squares-Discriminant Analysis enabled perfect discrimination of breast and thigh (100%), high accuracy for sex (> 89%), but limited discrimination for diet (< 67%). Overall, NIR spectroscopy demonstrated its potential as a rapid, non-destructive tool for predicting relevant chemical quality traits such as protein content and EE enhancing traceability in local poultry production systems.