An artificial neural network as a tool for kombucha fermentation improvement

Abstract

Kombucha as a tea-based fermented beverage has become progressively widespread, mainly in the functional food market because of health-improving benefits. As part of a daily diet for adults and children, kombucha stood out as a valuable non-alcoholic drink containing beneficial mixtures of organic acids, minerals, vitamins, proteins, polyphenols, etc. The influence of specific surface area of vessel, inoculum size, and initial tea concentration as operating parameters and fermentation time as output variable on the efficiency of kombucha fermentation was examined. The focus of this study is optimization and standardization of kombucha fermentation conditions using Box-Behnken's experimental design and applying an artificial neural network (ANN) predictive model for the fermentation process. The Broyden-Fletcher-Goldfarb-Shanno iterative algorithm was used to accelerate the calculation. The obtained ANN models for the pH value and titratable acidity showed good prediction capabilities (the r2 values during the training cycle for output variables were 0.990 and 0.994, respectively). Predictive ANN modelling demonstrated to be effective and reliable in establishing optimum kombucha fermentation process using selected operating parameters.