Machine learning versus laboratory metrics: A dual approach to fresh and frozen lamb meat assessment

Abstract

Accurate discrimination between fresh and frozen lamb meat is essential for maintaining quality control, consumer trust, and preventing fraud in the meat industry. This study assessed both destructive physicochemical analyses and non-destructive Near-Infrared (NIR) spectroscopy, combined with machine learning, for the effective classification of fresh and frozen lamb meat. Conventional laboratory measurements, including pH, color parameters (L*, a*, b*), lipid oxidation (TBARS), cooking loss, and Warner-Bratzler shear force, showed statistically significant differences (p < 0.05) between fresh and frozen samples, demonstrating their usefulness in quality evaluation. Although Principal Component Analysis (PCA) identified these physicochemical parameters as key discriminators, the limited data posed overfitting risks for classification models. To address this, NIR spectroscopy (700-1100 nm) was employed in conjunction with eight optimized machine learning classifiers. Five-fold cross-validation further confirmed the robustness of the spectral-based classification. Among them, XGBoost achieved the highest accuracy (91.2 %), precision (90.1 %), recall (92.6 %), F1 score (91.3 %), and ROC AUC (0.95), outperforming Random Forest and other models. The discriminatory performance remained stable when using only the 900-1100 nm region, demonstrating that accurate classification can be achieved with lowcost mini-spectrometers provided their spectral acquisition quality is sufficient. Learning curve analysis verified model generalizability, decision boundary visualization highlighted class separability, and feature importance ranking revealed key spectral and physicochemical attributes driving discrimination. These results highlight the complementary benefits of both approaches: physicochemical tests provide detailed and accurate analysis, albeit labor-intensive and destructive, whereas NIR spectroscopy, combined with machine learning, offers a rapid, reliable, and non-invasive alternative suitable for high-throughput industrial applications. This dual-method framework offers a promising pathway for robust and scalable authentication of fresh versus frozen lamb meat.