<![CDATA[The global obesity rate has tripled since 1975, driving the development of technology-based solutions for predicting body weight to mitigate disease risks. This study implements three models—Decision Tree Regressor, XGBoost Regressor, and Deep Learning—to project final body weight based on physiological variables (age, gender, BMR), nutritional factors (caloric intake, surplus/deficit), and lifestyle factors (physical activity, sleep, stress). The multidimensional dataset from community health posts includes TDEE calculations and BMR estimates using the Harris-Benedict Equation. Evaluation using RMSE and R² indicates XGBoost as the best-performing model (RMSE: 5.65; R²: 0.974), outperforming the Decision Tree (RMSE: 10.68; R²: 0.908) and Deep Learning (RMSE: 10.4; R²: 0.913) models. Key challenges include overfitting in the Decision Tree and Deep Learning's inability to capture outliers due to vanishing gradients. The analysis identifies energy balance, representation of extreme data, and regularization as critical factors for model stability. Hyperparameter optimization (learning rate, max\_depth) and data augmentation are recommended to enhance generalization. These findings offer an innovative framework for data-driven health technologies, reinforcing the role of artificial intelligence in precision public health interventions. Practically, the study advocates for the adoption of optimized predictive models integrating multidimensional variables for high accuracy, while highlighting the need for outlier handling and further clinical validation to ensure relevance in real-world scenarios.]]>
Copyrights © 2025
