COMPARATIVE STUDY OF MACHINE LEARNING ALGORITHMS FOR OBESITY RISK PREDICTION IN ADOLESCENTS

Abstract

Adolescent obesity has emerged as a serious public health concern due to its strong association with long-term health complications such as cardiovascular diseases, diabetes, and metabolic disorders. Early and accurate prediction of obesity risk can support timely preventive interventions and personalized healthcare strategies. This paper presents a comparative study of supervised and ensemble machine learning algorithms for multi-class obesity risk prediction using physiological, lifestyle, and behavioral data. A dataset comprising 2,111 instances with 17 attributes was utilized, including physical measurements, dietary habits, physical activity levels, family history, and sedentary behavior indicators. Comprehensive exploratory data analysis was conducted using diverse visualization techniques to validate feature relevance and uncover obesity-related patterns. Multiple machine learning models, namely Support Vector Machine, Random Forest, XGBoost, and a Stacking Classifier, were trained and evaluated for performance comparison. Experimental results show that ensemble-based models outperform traditional classifiers, with XGBoost and Stacking Classifier achieving superior predictive accuracy across seven obesity categories. Furthermore, the proposed approach was implemented as a web-based system supporting dataset upload, exploratory analysis, visualization, model training, real-time prediction, and report generation. The results demonstrate that ensemble learning combined with data-driven analysis provides a robust and effective framework for adolescent obesity risk prediction and can assist healthcare decision-making processes.