Preliminary results of implementing a machine-learning pipeline for predicting the risk of imminent osteoporosis fracture

This study presents a Machine Learning pipeline for predicting imminent fracture risk in osteoporotic patients using non-invasive clinical data. Based on a cleaned dataset of 3,718 records from the Ontario Osteoporosis Strategy, several ML models were evaluated. The Soft Voting classifier outperformed other approaches, achieving 76.0% accuracy, 74.0% F1-score, 76.0% recall, and 73.0% precision. Among individual models, the Support Vector Machine performed best with 70.8% accuracy. Key predictors included age, weight, and fall history. These results highlight the potential of Machine Learning to support cost-effective imminent fracture risk prediction and enable early, personalized interventions in osteoporosis care.