Enhancing Clinical Decision Support Systems with Explainable Machine Learning: An Interpretable Framework for Predictive

Abstract

The opacity of advanced machine learning models remains a significant barrier to their clinical adoption, despite their demonstrated potential to improve diagnostic accuracy. We present a framework for Clinical Decision Support Systems (CDSS) that integrates explainable artificial intelligence techniques to reconcile predictive performance with interpretability. Our study retrospectively analyzed approximately 5,000 patient records, incorporating demographic, physiological, lifestyle, and medical history variables. We implemented five machine learning models—Decision Tree, Random Forest, Logistic Regression, Explainable Boosting Machine, and XGBoost—with SHapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) providing post-hoc interpretability. The dataset was preprocessed with mean and mode imputation, then normalized using Min-Max scaling, and partitioned into training, validation, and testing subsets. Performance evaluation employed accuracy, precision, recall, and F1-score. The XGBoost model integrated with SHAP achieved the highest accuracy of 94.1%, with 93.7% precision, 94.0% recall, and 93.8% F1-score, while Logistic Regression—the most interpretable model—yielded 85.4% accuracy. Feature importance analysis identified blood glucose, cholesterol, age, BMI, and blood pressure as the dominant predictors. SHAP summary plots revealed that elevated glucose and cholesterol levels as strong contributors to increased disease risk, while LIME explanations provided transparent reasoning for individual predictions. Our contribution is a comparative framework demonstrating that high predictive accuracy and clinical transparency are not mutually exclusive. This framework thus bridges the gap between complex computational methods and practical healthcare deployment, supporting evidence-based, patient-centered care through understandable diagnostic insights.