An electronic sense-based machine learning model to predict formulas and processes for vegetable-fruit beverages

Developing vegetable-fruit beverages is complicated, requiring redundant formula design and continuous process adjustment and hindering rapid intervention. This study aimed to solve the problem of high cost and long cycle of traditional research and development mode in beverage demand diversification by developing a model combining electronic sense and machine learning to predict formulas and processes for vegetable-fruit beverages. In this study, two novel data processing methods of data fusion and data imputation were applied to increase the scope for further model analyses. The k-nearest neighbour (kNN), random forest (RF), and deep neural network (DNN) were used to predict the formulas and processes of sea buckthorn-passion fruit juice beverages based on the electronic sense. The DNN model showed the best performance compared with the other two models, with an R2 of 0.88. The external validation indicated the excellent robustness of the predictive model. Overall, the model established successfully and could predict formulas and processes of vegetable-fruit beverages and provide theoretical and practical support for beverage development.