Application of Machine Learning Algorithms for Obesity Prediction and Identification of Influencing Factors Using SHAP

Abstract

Obesity has become a serious global health problem due to lifestyle changes from traditional lifestyles to sedentary lifestyles characterized by low physical activity and high-calorie, low-fiber diets. This condition is exacerbated by various factors, such as gender, family history, physical activity, and calorie intake, which collectively increase the risk of obesity in adolescents and adults. Therefore, early obesity prediction efforts are needed to support more effective preventive decision-making. This study aims to develop an obesity prediction model using Decision Tree, Random Forest, CatBoost, and Artificial Neural Network (ANN) algorithms to predict obesity probability and find the best method from the four algorithms. As a result of this analysis, obesity classes were predicted with success rates of 77.00%, 85.00%, 86.00%, and 84.00%, respectively. CatBoost was the most successful method for this dataset and classified obesity with a success rate of 86.00%. Analysis using SHAP values revealed that the features Food Intake Between Meals and Consumption of Fast Food had the greatest influence in increasing the probability of overweight/obesity. Conversely, features such as Physical Exercise and Height contributed negatively to the probability of obesity.