M.S. Applied Data Science - Capstone Chronicles 2025

3

associations between lifestyle behaviors and metabolic health, few studies have evaluated how well these indicators can classify risk using machine learning—particularly in comparison to models that include pharmaceutical variables. 2.1 Problem Identification and Motivation While pharmaceutical treatment is an important component of managing chronic metabolic disease, its widespread use has also fostered a passive approach to health: treating symptoms rather than addressing root causes. Many individuals are placed on long-term medications without parallel efforts to change lifestyle habits. This is not only unsustainable from a public health and cost standpoint but may also reduce patient agency in long-term health outcomes. Additionally, clinical decision-making often focuses on prescribing medications based on lab values without a full evaluation of behavioral context. This research is motivated by a growing recognition that many chronic conditions—particularly obesity—may be reversible or preventable through behavioral change. Still, the extent to which lifestyle alone can drive positive outcomes remains a topic of debate, particularly given variation in individual behaviors. By quantifying the predictive value of lifestyle and behavioral indicators—with and without medication data—this project seeks to bridge a gap in how we evaluate and manage chronic metabolic conditions from a data science perspective. 2.2 Definition of Objectives This project has three main objectives. First, it aims to build machine learning classification models to predict whether an individual has metabolic syndrome, using only lifestyle and behavioral indicators such as diet, physical

activity, and other non-clinical habits. Second, it compares these models to ones that include pharmaceutical use as a predictor to evaluate whether medication contributes significant predictive value beyond lifestyle features. Third, the project analyzes model outputs to identify which lifestyle and behavioral features are most influential in predicting metabolic syndrome status. If models using only lifestyle and behavioral data perform comparably to those that include medication, this would support prevention-oriented strategies focused on modifiable behaviors to identify individuals at risk. On the other hand, if including medication use significantly improves predictive accuracy, it may reflect both the therapeutic effects of pharmaceuticals and their role as a proxy for prior diagnosis or disease severity. In such cases, strong predictive power from medication variables could suggest that these individuals were identified earlier or are experiencing more advanced stages of the condition—emphasizing the need for earlier detection and targeted intervention. These findings can help inform public health efforts by clarifying when lifestyle-based prevention may be sufficient and when clinical management is likely necessary. 3 Literature Review To understand the predictive power of lifestyle factors in metabolic health outcomes and the role of medication in such models, it is essential to position this work within current scientific literature. This review draws on recent empirical and clinical studies focused on obesity treatment, machine learning applications in metabolic

154