The Role Of Physical Activity In Obesity Prediction: Insights From Machine Learning Models

PURPOSE: The purpose of this study was to examine the role of physical activity as a predictor of obesity by using machine learning models. By integrating multiple data types, including physical activity parameters, demographic, dietary, and health characteristics, we aimed to assess the effectiveness of physical activity indicators in predicting obesity status. METHODS: We applied three machine learning classifiers—Random Forest, Naive Bayes, and Gradient Boosting Machine (GBM)—to a dataset containing physical activity, demographic, dietary, and health data. The classifiers were trained and evaluated to determine their accuracy and ability to predict obesity. Cross-validation and test accuracy metrics, including AUC-ROC and F1-weighted scores, were used to evaluate the performance of each model. RESULTS: The Random Forest classifier achieved the highest performance, with a cross-validation accuracy of 95.26%, test accuracy of 97.88%, and an AUC-ROC score of 1.00. Physical activity features demonstrated a high F1-weighted score of 0.98, indicating that activity level patterns were closely associated with obesity status. The Naive Bayes classifier, though simpler, also performed well, achieving 93.93% cross-validation accuracy, 93.23% test accuracy, and an AUC-ROC score of 0.98. In contrast, the Gradient Boosting Machine (GBM) had significantly lower performance, with a test accuracy of 70.72% and high rates of misclassification for obesity cases. Feature analysis revealed that general activity indicators related to vigorous and moderate physical activities were critical predictors of obesity. CONCLUSIONS: This study demonstrated that integrating physical activity data into machine learning models effectively identifies individuals at risk of obesity, emphasizing physical activity as a key predictor. The Random Forest classifier outperformed other models, supporting its suitability for predicting obesity based on activity patterns. Future research should explore incorporating wearable device data for real-time monitoring and personalized interventions to enhance obesity prediction and prevention strategies.