NIRS Coupled With Machine Learning Algorithms for the Authentication and Quality Assessment of Indigenous Cattle and Buffalo Meat

ABSTRACT Ensuring the authenticity of high‐value indigenous meats such as cattle ( Bos indicus ) and water buffalo ( Bubalus bubalis ) has gained heightened attention as a result of frequent adulteration in global supply chains. This study established a rapid, non‐destructive authentication framework comparing physicochemical profiling and portable near‐infrared spectroscopy (NIRS) combined with machine learning (ML). Forty longissimus thoracis et lumborum samples ( n = 20/species) were analyzed for pH, color ( L *, a *, b *), water‐holding capacity, drip loss, cooking loss, and shear force. A total of 4000 NIR spectra (700–1100 nm) were acquired using a handheld spectrometer. Buffalo samples exhibited significantly higher shear force (77.82 N), cooking loss (32.56%), and redness ( a * = 20.92) than cattle ( p < 0.001). NIRS analysis revealed distinct absorbance peaks in buffalo in the O–H (900–980 nm) and C─ H/N─H (750–850, 1060–1088 nm) regions. ML models, including XGBoost and a 1D convolutional neural network, achieved 98.12% and 98.4% accuracy, respectively. SHAP analysis identified 1075.30 nm as a key differentiating wavelength associated with protein and moisture. However, further validation using larger and more diverse sample sets, external test datasets, and intentional adulteration scenarios is required before confirming its broader applicability for routine authentication and fraud detection.