Can We Predict Older Adults’ Long-Term Healthy Behavior Maintenance? Identifying the Optimal Machine Learning Model

As population aging accelerates, maintaining long-term health behaviors has become a major challenge in older adult health management. To address the limitation that prior studies emphasize short-term behavior adoption, this study proposes a predictive framework integrating Health Belief Model (HBM) psychological variables with mobile health (mHealth) usage behaviors. By comparing Random Forest, XGBoost, and AdaBoost, the study predicts whether older adults can enter and sustain the health behavior maintenance stage and examines the relative importance of psychological, behavioral, and demographic factors to support intelligent, personalized health promotion. Using data from 1,500 adults aged 60 and above, results show that combining HBM factors with mHealth usage features enables effective prediction, with Random Forest performing best. Self-efficacy, perceived severity, perceived susceptibility, cues to action, and mHealth usage frequency emerged as key predictors, while demographic variables played a minor role, offering practical guidance for targeted digital health interventions.