Machine learning-driven optimization of Riceberry donut formulations for high sensory acceptance and low calorie

This study explores the enhancement of Riceberry donut formulations using Random Forest Regression (RFR) to develop recipes that are both nutritious and sensory appealing. A collected dataset has been utilized, highlighting crucial components such as component ratios (e.g., Wheat Flour, Riceberry Flour, Shortening, Egg), physical features (e.g., texture, wetness, chewiness), and nutrient composition (e.g., carbohydrates, lipids, proteins). The model was developed to forecast four principal output variables: Overall Liking, Taste, Texture, and Calories (kcal). The results exhibited great predictive accuracy, with predicted values closely matching real values for all sensory qualities, underscoring the model’s robustness in reflecting consumer opinions. The experiment revealed that Riceberry flour considerably challenged the composition of nutrients and sensory features of the donuts. Employing the enhanced model, simulated recipes were generated by adjusting Riceberry flour (20-35%), shortening (15-19g), and protein (2.6-3.0g), and the most effective formulation, consisting of 30% Riceberry flour, 15g of shortening, and 2.7g of protein, obtained a predicted Overall Liking score of 7.37 and 114.5 kcal per 30g serving. The results presented underscore the potential advantages of Riceberry flour in producing health-focused baked goods that are also attractive to customer preferences, illustrating the efficacy of machine learning in balancing nutritional quantity and sensory pleasure in food composition.