A Data-Driven Machine Learning Framework for Predicting Disabilities in Cerebral Palsy

The use of artificial intelligence (AI) in medicine and healthcare has expanded significantly, particularly in classification and prediction tasks. Machine learning algorithms analyze medical data, identify patterns, and generate predictions that assist in clinical decision-making. However, challenges such as data quality, redundancy, and sensitivity hinder the development of optimal predictive models. This study focuses on improving impairment level prediction by leveraging high-quality datasets and advanced machine learning techniques. Using medical records from cerebral palsy patients, we evaluate six machine learning algorithms: Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest (RF), Logistic Regression (LR), Naïve Bayes (NB), and XGBoost. Our objective is to determine the most effective model for predicting Gross Motor Function Classification System (GMFCS) levels, a key indicator in cerebral palsy assessment. Additionally, we introduce MEDX-Ensemble, a novel interpretable ensemble model that integrates multiple classifiers to enhance prediction accuracy while maintaining clinical explainability. Performance evaluation, including Weighted Kappa scores, reveals that advanced models like XGBoost and MEDX-Ensemble achieve the highest accuracy and robustness, demonstrating substantial agreement with physician classifications. These results demonstrate the potential of AI-driven models to enhance patient assessment and clinical decision-making. Through rigorous validation of diverse machine learning techniques, this study identifies optimal methodologies for GMFCS level prediction, contributing to reliable and interpretable healthcare predictive models.