An XGBoost-Based Model with SHAP Explainability for Muscular Dystrophy Classification Using Synthetic Clinical Data

Purpose of the study: Muscular dystrophies (MDs) are uncommon, progressive genetic conditions that result in muscle degeneration. This study aims to apply an interpretable machine learning approach using XGBoost to classify muscular dystrophy subtypes from synthetic clinical data. Methodology: We implemented an XGBoost model with engineered features such as interaction terms, trained on 1,000 synthetic records containing demographic, clinical, and genomic-like variables. SHAP (SHapley Additive exPlanations) was used for post-hoc interpretability of the predictions. Main Findings: The proposed model achieved high accuracy (92.4%) and macro F1-score (91.3%) across muscular dystrophy subtypes. SHAP analysis confirmed that features such as inheritance pattern, primary muscle group, creatine kinase levels, and severity score were the most influential. Applications of the study: The framework may serve as a diagnostic support system in clinics, telemedicine, and trial triaging for neuromuscular disorders. Its explainability ensures usability in medical decision-making. Novelty/Originality of the study: This work combines XGBoost with SHAP explainability specifically for muscular dystrophy classification, providing interpretable AI insights using synthetic data for privacy and scalability.