ADS Capstone Chronicles Revised

28

accounts for a portion of the variance, while negative values indicate poor alignment with the data. 5.1.1 Combined Results: Menu and Individual Food Data The regression results are summarized in Table 5, showing the performance of Linear Regression, Random Forest Regressor, XGBoost Regressor, and Support Vector Regressor (applied only to individual food data). Across both datasets, Linear Regression demonstrated the lowest MSE, making it the most accurate model in predicting patient scores. However, negative R-squared values were observed in all models, indicating challenges in capturing variability within the data. Regression models applied to menu food data demonstrated low MSE values, with the Linear Regression model achieving the lowest MSE at 0.027, closely followed by XGBoost at 0.028. These values suggest small differences between predicted and actual patient scores. Despite this, the models failed to capture variability effectively, as evidenced by negative R-squared scores across the board. This is likely due to the homogeneous nature of menu items, which

often share ingredients and nutritional profiles within the same restaurant or across similar establishments. Consequently, the models struggled to account for distinctions in patient scores that might arise from subtle differences in these features. For individual food data, the regression models displayed clearer performance distinctions. Linear Regression again emerged as the best-performing model, with the lowest MSE at 0.044, indicating a strong ability to predict patient scores. XGBoost and Random Forest followed with slightly higher MSE values of 0.049 and 0.048, respectively. The Support Vector Regressor, included due to its effectiveness on smaller datasets, achieved an MSE of 0.050. As with the menu food data, all models reported negative R-squared values for individual food data. This outcome reflects the limited variability in the nutritional content of individual food items, which constrains the models' ability to fully explain differences in patient scores. However, the low MSE values indicate that the models are making accurate predictions despite this limitation.

Table 5 Regression Model Analysis for Menu and Individual Food Datasets

230