ADVANCED MACHINE LEARNING APPROACHES FOR PREDICTIVE MODELING OF MICROBIAL GROWTH DYNAMICS AND SHELF LIFE IN FUNCTIONAL FOODS

Abstract

The escalating global demand for functional foods—products formulated to confer health benefits beyond basic nutrition—has intensified the need for robust predictive frameworks capable of ensuring microbiological safety and quality throughout complex supply chains. Traditional predictive microbiology models, while providing foundational contributions to the field, exhibit inherent limitations in capturing the intricate, nonlinear interactions among intrinsic food properties, extrinsic environmental factors, and microbial population dynamics within functional food matrices. This investigation presents a comprehensive evaluation of advanced machine learning (ML) methodologies for predictive modeling of microbial growth kinetics and shelf life estimation in functional foods. Five ML algorithms—Radial Basis Function Neural Networks (RBF-NN), Random Forest Regression (RFR), Support Vector Regression (SVR), Gaussian Process Regression (GPR), and Long Short-Term Memory (LSTM) networks—were systematically assessed against conventional primary growth models (Gompertz, Logistic, and Baranyi) utilizing a multi-dataset framework encompassing diverse functional food matrices. Performance metrics including adjusted R², Root Mean Square Error (RMSE), Bias Factor (Bf), and Accuracy Factor (Af) revealed that ML approaches consistently surpassed traditional models, with GPR achieving the highest predictive accuracy (R²adj = 0.959, RMSE = 0.153) and LSTM demonstrating superior capacity for capturing temporal dependencies under dynamic temperature conditions (72.5% RMSE reduction compared to Gompertz). The RBF neural network model exhibited exceptional performance in shelf life prediction with a relative error of 2.66%, substantially outperforming conventional kinetic models. SHapley Additive exPlanations (SHAP)-based interpretability analysis identified temperature, pH, and water activity as the most influential predictors across all models. These findings establish machine learning as a transformative paradigm in predictive food microbiology, offering unprecedented accuracy flexibility, and data-driven insight for shelf life estimation in functional foods, with significant implications for food safety management and quality assurance in the functional food industry.