Machine Learning Modeling For Predicting Adherence To Physical Activity Guideline

As regular physical activity (PA) has been confirmed to provide significant health benefits, several PA guidelines were developed to improve individuals’ health. Given the significance of adhering to PA guidelines, numerous studies have been conducted to predict and understand factors influencing adherence to these guidelines. Machine Learning (ML) classification has become a widely adopted technique, surpassing classical analysis methods in performance. Despite its proven efficacy, only a few studies have employed ML for classifying individuals' PA based on comprehensive predictors. PURPOSE: To create predictive models for PA guidelines by using ML and examine the critical determinants influencing adherence to the PA guidelines. METHODS: Utilizing data from 2009-2018 National Health and Nutrition Examination Survey, 13,699 entries were analyzed after excluding individuals under 18 (n = 18,761), those with physical limitations in PA (n = 14,770), and missing data (n = 2,463). Variables were classified into demographic, anthropometric, and lifestyle categories. Using six machine learning, which are logistic regression, Support Vector Machine (SVM), decision tree, random forest, eXtreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (Light GBM), 42 prediction models were created across seven category combinations (e.g., all, demographic and lifestyle, demographic and anthropometric, and demographic). Model performance was evaluated via accuracy, precision, recall, and F1 score. Permutation feature importance assessed variable significance in each model. RESULTS: Among the array of algorithms considered, SVM using a dataset of demographic and lifestyle categories emerged as the most effective method in the prediction for PA guidelines (accuracy = .708, precision = .715, recall = .961, and F1 score = .820). Additionally, based on the PFI, sedentary behavior (PFI = .018) was the most important variable in the top models followed by age (PFI = .011), education status (PFI = .010), and gender (PFI = .006). CONCLUSION: These results highlight the possibilities of using data-driven methods with ML in PA research. Our analysis also identified crucial variables, providing valuable insights for targeted interventions aimed at enhancing individuals' adherence to PA guidelines.