ADS Capstone Chronicles Revised

1

Smart Meal Choices A Data Science Approach to Personalized Diabetes-Friendly Restaurant Meal Recommendations

Claire Bentzen Applied Data Science Master’s Program Shiley Marcos School of Engineering / University of SanDiego cbentzen@sandiego.edu

Tara Dehdari Applied Data Science Master’s Program Shiley Marcos School of Engineering / University of SanDiego tdehdari@sandiego.edu

Logan Van Dine Applied Data Science Master’s Program Shiley Marcos School of Engineering / University of SanDiego lvandine@sandiego.edu

ABSTRACT This study aims to provide support for diabetes management by establishing a personalized food recommendation system that streamlines the process of making informed dietary decisions at restaurants and maintaining health goals. Using restaurant menu data, general food nutritional information, and individual patient health data, various methods–including linear regression, random forest, and XGBoost–were evaluated to determine the most optimized model for delivering personalized food recommendations. Two approaches were employed: a regression analysis to predict the suitability score of a food item for an individual, and a classification analysis to provide a user-friendly deployment of the recommendations. Following cross validation and hyperparameter tuning, XGBoost emerged as the best performing model for both regression and classification, based on accuracy, precision, recall, R2 score, and interpretability. The most important features in this model are body mass index (BMI), general suitability score, high cholesterol, and sugars. This analysis extends standard dietary recommendations by incorporating a hybrid

approach that tailors suggestions to the unique needs of individuals with diabetes. Interpretable and actionable recommendations are achieved by aligning personal suitability scores with general dietary suggestions, emphasizing important elements such as sugar and carbohydrates. Overall, this study offers a unique solution to diabetes management by empowering individuals to make informed dining choices. KEYWORDS diabetes, Type 1, Type 2, recommendation system, nutrition, glycemic load, carbohydrate count, insulin, glucose management, classification, regression 1 Introduction Diabetes is a nationwide epidemic that affects 11.6% of the U.S. population, including more than 300,000 children who may be too young to recognize optimal diet choices for their glucose levels (Centers for Disease Control and Prevention, 2024). This disease can present in two forms: Type 1 and Type 2 (American Diabetes Association, 2024). As the eighth leading cause of death in the United States, prevention and management support is critical. Treatment plans involve a combination of blood

203