Optimization and Process Validation of Freeze-Structured Meat Substitute Using Machine Learning Models

Abstract

The increasing demand for sustainable and healthy food sources has catalyzed the rapid expansion of the meat substitute market. Legumes have been offering significant nutritional advantages and making them a superior choice for meat substitutes due to their advantages in protein quality. Legume composites such as pea protein isolate (PPI) and isolated soy protein (ISP) and vital wheat gluten combine with other ingredients to create a fibrous texture. A mixture design approach was employed to optimize the formulation with freeze structuring methodology; machine learning (ML) models were integrated to predict the textural and sensory properties of the resulting meat analogues. The final samples were tested for textural, water activity, sensory properties, and water activity, which is used to validate the consistency of the experimental formulations. The optimized formulation, consisting of 18.9% PPI, 12.6% ISP, and 8.5% VWG, exhibited a hardness of 815.587 N, springiness of 0.845 mm, overall acceptability of 8.7, and water activity of 0.887. Integrating five ML algorithms for built-in feature selection and classification mechanisms predicted the desired properties. The predicted values from the ML models closely matched the experimental results, demonstrating the potential of this approach with a negligible amount of difference in experimental values. Among the ML models tested, Gradient Boosting provided the best prediction for hardness (RMSE = 24.698, R² = 0.986), AdaBoost performed best for springiness (RMSE = 0.019, R² = 0.940) and overall acceptability (RMSE = 0.284, R² = 0.904), while XGBoost showed the highest accuracy for water activity prediction (RMSE = 0.002, R² = 0.985). In conclusion, the integration of both approaches emphasizes the importance of reducing dimensionality and enhancing data quality, and this research serves as a platform for future work studies in the field of plant-based meat alternatives.