M.S. Applied Data Science - Capstone Chronicles 2025

4

prediction, and comparative effectiveness of pharmaceutical and lifestyle interventions. The review is structured thematically to reflect five major areas of relevance. 3.1 Effectiveness of Medication versus Lifestyle Interventions Mechanick et al. (2023) provide a detailed review of the impact of second-generation anti-obesity drugs, such as GLP-1 receptor agonists (e.g., semaglutide), showing up to 15–21% weight loss in clinical populations. They contrast this with traditional lifestyle interventions that typically result in 5–10% weight loss over the same period. While medications are effective in the short term, the study notes that long-term success still depends on sustained behavioral change. This work validates the inclusion of pharmaceutical variables in prediction models while reinforcing that medication alone is insufficient for sustainable results. The research gap lies in quantifying this relationship in real-world populations using predictive models like those proposed in this study. These findings are consistent with broader epidemiological trends, as a systematic review of the Asia–Pacific region reported rising prevalence of metabolic syndrome and underscored regional variation in associated risk factors (Sigit et al., 2020). 3.2 Impact of Lifestyle-Only Interventions on Metabolic Health Jensen et al. (2019) reviewed over two decades of behavioral intervention research and confirmed that structured lifestyle programs (e.g., dietary counseling, physical activity plans) lead to significant improvements in metabolic health. The average participant lost about 8% of their baseline weight within six months, with blood pressure and glucose metrics also improving. Similarly, a systematic review and meta-analysis of prospective cohort studies found that lifestyle

risk factors—including poor diet, physical inactivity, and smoking—were strongly associated with increased incidence of metabolic syndrome (Lotfaliany et al., 2021). This research aligns with the hypothesis that lifestyle factors have measurable impacts on metabolic health and supports the exploration of their standalone predictive power. However, while this work identifies effect, it does not explore prediction, which is where this project contributes new value. 3.3 Machine Learning Approaches Using NHANES Data Huang and Huang (2024) applied XGBoost and SHAP to predict obesity status using NHANES 2017–2020 data. Their models achieved high accuracy (AUC ≈ 0.83), identifying variables like waist circumference and HDL as top predictors. This study validates the technical feasibility of building interpretable ML models using NHANES and supports the current project's approach. However, their work did not examine the role of medication use in prediction nor focus on the comparative strength of lifestyle features alone, leaving room for this study to extend their methodology with a sharper public health focus. 3.4 Interpretable ML Models of Lifestyle Factors and Obesity Sun et al. (2024) used NHANES and CHNS data to model obesity using gradient boosting machines, showing that alcohol intake, sedentary behavior, and dietary protein intake were highly influential predictors. Their use of SHAP values helped identify non-obvious relationships and demonstrated the potential of interpretable models in public health research. This aligns with the current project's emphasis on transparency and interpretability but differs in scope, as Sun et al. did not include or compare the contribution of

155