Predictive model establishment for forward-head posture disorder in primary-school-aged children based on multiple machine learning algorithms.

Background: Forward head posture frequently occurs among primary school children, potentially due to prolonged sedentary behavior associated with academic demands and reduced physical activity. However, existing prevention and screening methods fail to accurately and promptly predict the onset of forward head posture.

Objective: This study aims to identify highly sensitive predictive indicators for forward head posture in primary school children using the Least Absolute Shrinkage and Selection Operator (LASSO) regression algorithm. Multiple machine learning algorithms are applied to construct distinct risk prediction models, with the most effective model selected through comparative analysis. The Shapley Additive Explanations (SHAP) method is used to quantify the influence of each feature on model outcomes, ensuring enhanced model interpretability.

Methods: Employing a cross-sectional study design, this research recruited 520 primary school-aged children, gathering data on demographics, anthropometrics, and physical activity levels. Univariate logistic regression was utilized to identify high-risk factors for forward head posture. The LASSO algorithm was subsequently applied to select key predictors. Six machine learning models-K-nearest neighbor (KNN), light gradient boosting machine (LGBM), extreme gradient boosting (XGBoost), random forest (RF), linear model (LM), and support vector machine (SVM)-were developed to predict risk. The performance of each model was evaluated, and the best-performing model was further interpreted using the Shapley Additive Explanations (SHAP) algorithm.

Results: A total of 514 children were ultimately included in the study, of whom 300 exhibited forward head posture. LASSO analysis identified age, bodyweight, BMI, sex, and weekly total homework time as prominent risk indicators. Among the 6 predictive models, the random forest algorithm demonstrated the highest performance (AUC = 0.865), significantly outperforming the others. SHAP analysis revealed that BMI, bodyweight, and age were the most influential predictors, with BMI contributing the most.

Conclusion: The random forest-based prediction model achieved superior predictive accuracy for forward head posture among Chinese primary school children, emphasizing the importance of monitoring BMI, bodyweight, and age for early intervention and prevention efforts.