A machine learning-based framework for individualized physical activity and strength targets related to women’s bone health

While substantial evidence supports the associations between physical activity and bone health, the present study aims to advance the field by applying a novel, data-driven approach through machine learning techniques. The primary objective was to identify significant predictors of femoral neck bone mineral density (BMD) and to develop an interpretable modeling framework for estimating individualized physical activity and strength targets associated with BMD in women. This study analyzed data from 1205 female participants in the National Health and Nutrition Examination Survey (NHANES) 2013–2014 cohort to examine the relationship between femoral neck BMD and key variables, including age, body mass, grip strength, and physical activity. Machine learning regression models, such as Multiple Linear Regression (MLR) and Random Forest Regression (RFR), were employed to develop an interpretable modeling framework. Shapley Additive exPlanations (SHAP) were used to support interpretation of the nonlinear RFR model and to visualize each predictor’s contribution to BMD estimates. The MLR model estimated femoral neck BMD with a test R ² of 0.375, RMSE of 0.105 g/cm ² , and MAE of 0.082 g/cm ² . The RFR model resulted in a test R ² of 0.309, RMSE of 0.111 g/cm ² , and MAE of 0.086 g/cm ² . This study illustrates the potential of interpretable machine learning approaches to support individualized, data-driven assessment of physical activity, strength, and BMD. The numerical model developed in this work provides association-based estimates of physical activity and strength targets according to individual characteristics, highlighting its potential for hypothesis generation and future studies on bone health promotion and osteoporosis prevention. Prospective and interventional validation is required before clinical recommendations can be made.