Machine Learning-Based Model for Grip Strength Prediction in Healthy Adults: A Nationwide Dataset-Based Study

Background: This study aimed to develop machine learning models for estimating the handgrip strength (HGS) in healthy adults and to identify the model with the highest accuracy and generalizability. Methods: Data from the Korean National Health and Nutrition Examination Survey (2014-2019), including 21,147 participants aged >19 years, were analyzed. The maximum HGS was measured using a standardized protocol, with 11 demographic, anthropometric, and physical activity predictors. Polynomial regression (PR), multilayer perceptron (MLP), and extreme gradient boosting (XGBoost) models were developed and evaluated using the root-mean-square error (RMSE) and coefficient of determination (R²). Results: The HGS was found to vary by gender, age, and hand dominance, with males and younger individuals showing higher values. The XGBoost model achieved the highest R² (0.717), demonstrating superior predictive accuracy and generalizability compared with PR and MLP. Key predictors in the XGBoost model included weight, age, height, and waist circumference, while hand dominance was less significant. Conclusions: The XGBoost model outperformed the MLP and PR models, achieving the highest R² value. It holds promise for clinical applications, enabling accurate HGS estimation to support early diagnosis, targeted interventions, and personalized goal setting.