mrselemogo777-gif/ML-Assisted-Fracture-Triage-for-Rural-African-Clinics-A-Comparative-Study-

# ML-Assisted Fracture Triage for Rural African Clinics: A Comparative Study ## Abstract This study compares clinical, X-ray, and multimodal ensemble models for fracture triage using real-world data from 1,129 patients across 18 African countries. We demonstrate that a simple clinical model (Random Forest) achieves 81.4% accuracy using only patient vitals, while an XGBoost ensemble combining clinical and X-ray data reaches 82.4% accuracy. ## Model Performance | Model | Accuracy | |-------|----------| | Clinical (Random Forest) | 81.4% | | X-ray (EfficientNetB0 + RF) | 67.7% | | Multimodal Ensemble (XGBoost) | 82.4% | ## Dataset - 1,129 patients - 18 African countries - 10 fracture types (5 Simple, 5 Complex) - 790 training X-rays, 226 test X-rays ## Cross-Validation - Leave-one-country-out CV: 83.8% (±5.0%) ## Repository Structure - `african_fracture_ai_notebook.ipynb` – Complete training and evaluation pipeline - `requirements.txt` – Python dependencies - Webapp available at: mrselemogo777-gif/african-f… ## Citation If you use this work, please cite: Selemogo, L. (2026). ML-Assisted Fracture Triage for Rural African Clinics: A Comparative Study.