Unsupervised Learning for Optimal Personalized Dietary Menus to Prevent Diabetes and Cardiovascular Diseases

Healthy diets can slow disease progression, but their effectiveness may decrease. Patients often give up these diets due to limited food choices, unappetizing meals, and reduced physical activity from cutting calories. To address this, we developed an intelligent nutritional balance system to prevent cardio-diabetic diseases. This system creates diets that optimize cholesterol and glycemic control through the following steps: (a) Characterizing Moroccan foods based on 19 nutrients and their glycemic load, (b) Classifying foods using a Gaussian mixture model, (c) Modeling the optimal diet with a fuzzy mathematical model using recommendations from the WHO, USDA, and FAO, (d) Solving the model with a genetic algorithm, (e) Translating portions and food groups to meet constraints, and (f) Resolving the final model using the backtracking method. We implemented this strategy based on the main foods consumed in Morocco, considering different levels of belief (0.25, 0.5, 0.75) regarding the glycemic load of these foods. The results show that the custom artificial diets align with WHO, USDA, FAO, and DGA recommendations. The menus are flexible, allowing for substituting expensive or rare foods with more affordable and readily available alternatives without compromising the quality of the diets.