Modelling and optimization of synbiotic yogurt fermentation and development of machine learning-based fermentation software

Abstract

The fermentation behaviour of Lactobacillus delbrueckii subsp. bulgaricus, Streptococcus thermophilus, and Lactobacillus acidophilus during synbiotic yogurt production enriched with freeze-dried banana powder was investigated in this study with the aim of characterizing microbial interactions, optimizing growth conditions, and developing a machine learning-based tool to predict the fermentation process. Fermentation experiments were performed under controlled temperature (40–43 °C), and prebiotic concentrations (0–30 g·l-1). Machine learning algorithms, including decision tree regression (DTR), random forests (RF), and Gaussian process regression (GPR), were then employed to predict bacterial growth behaviour and identify optimal fermentation conditions. GPR provided an adjusted coefficient of determination (R2adj) value greater than 0.82 for each bacterial strain, making GPR the best machine learning model. Therefore, it was integrated into the software to create a virtual simulation environment capable of representing fermentation dynamics. Based on modelling, the optimum fermentation conditions were determined as 41.5 °C and 20 g·l-1 freeze-dried banana powder concentration. Overall, the findings demonstrate that banana powder boosts probiotic growth, and machine learning enables efficient optimization and monitoring of the synbiotic yogurt fermentation process.