PREDICTING THE TENDERNESS OF BROILER BREAST FILLETS USING VISIBLE/NEAR-INFRARED SPECTROSCOPY AND MACHINE LEARNING ALGORITHMS

Chicken, as the most consumed meat worldwide, plays a crucial role in daily protein nutrition. Typically, consumers choose meat based on its tenderness, one of the quality traits in meat selections. In general, assessing the tenderness of chicken meat is commonly labour-intensive and destructive. An alternative approach to traditional methods is NIR spectroscopy, which is non-invasive, rapid, and economical. Portable NIR devices enable real-time, non-destructive meat quality assessment. Hence, this study investigates the effectiveness of a portable NIR spectroscopy system in predicting chicken meat shear force by comparing several machine learning algorithms, which are linear model (PCR) and non-linear model (ANN and PCR-ANN). The result shows prediction accuracy, Rp of 0.64, 0.73, and 0.91 for PCR, ANN, and PCR-ANN models, respectively. The calibration sets, Rc, for all models are 0.65, 0.73, and 0.92 for PCR, ANN, and PCR-ANN, respectively. The PCR-ANN model successfully outperforms both PCR and ANN, indicating it is amenable to non-invasive shear force prediction applications. In conclusion, the PCR-ANN model sufficiently achieved the best performance in predicting chicken meat tenderness using near-infrared spectral data. This study is essential for a number of reasons, including sustaining consumer satisfaction in the food industry, increasing industry returns, and increasing eating quality.