ADS Capstone Chronicles Revised

35

6 Discussion This data science objective was focused on creating a personalized food recommendation system for diabetic individuals by incorporating nutritional data from restaurants and individual food items with simulated patient data. The project demonstrated that XGBoost was most effective in predicting binary recommendations for diabetic-friendly meals. The results highlighted key findings about the systems performance. When compared with previous studies, the project extends existing findings by combining health-specific scoring with machine learning techniques tailored for diabetic individuals. While prior research focused on general restaurant or recipe recommendations, the project incorporates glycemic management needs into a hybrid recommendation model. This combination of data allows for the creation of a system that can be deployed and used by individual diabetic patients. To identify the best model, classification techniques to categorize foods as “recommended” or “not recommended” and regression techniques to predict patient scores were employed. They were used to assess the suitability of various foods. The XGBoost classifier showed the best results by achieving great performance across both individual food and restaurant datasets. The high precision scores, 0.994 for menu data and 0.989 for individual food data, indicates that the model effectively minimizes false positives which is an important requirement for health-related recommendations. Similarly, the regression models showed low mean squared error (MSE), with Linear Regression performing best in predicting patient scores for both menu and individual data. However, negative R-squared values across regression models highlight the

5.5.3 Cross-Validation Cross-validation confirmed the robustness of the XGBoost model. The 5-fold cross-validation results for menu recommendations (Table 9) and individual food recommendations (Table 10) demonstrated consistently high accuracy across all folds, with minimal variance, underscoring the model's stability and generalizability. These results underline the stability and generalizability of XGBoost across different subsets of the data. Table 9 Cross-Validation Results for Menu Recommendations

Table 10 Cross-Validation Results for Individual Foods

237