Prediction of Cardiovascular Risk Using Machine Learning Based on Maximal Oxygen Consumption , Physical Fitness, and Anthropometry

) cut-points. Data were standardised, class imbalance was addressed using Synthetic Minority Over-sampling TEchnique (SMOTE), and eight supervised classifiers were trained with stratified five-fold cross-validation and grid search. Ensemble tree-based methods outperformed kernel-based models. Gradient Boosting achieved the best balance between predictive performance (area under the curve-receiver operating curve AUC-ROC 0.716, F1-score 0.857, recall 0.760, accuracy 0.755), followed by Random Forest. SHapley Additive exPlanations (SHAP) analyses indicated that muscular fitness measures, particularly push-ups, sit-ups and standing long jump, contributed most to risk classification, whereas anthropometric indicators showed lower importance. These findings suggest that machine learning models built from school-based muscular fitness tests and basic anthropometry can discriminate adolescents at cardiovascular risk due to low cardiorespiratory fitness, offering a feasible, low-cost strategy to support early identification and targeted physical activity interventions.