Hybrid Interpretability and Performance Healthcare Model

Abstract

The development of AI models in healthcare AI is hampered by a central dilemma of improving both requirements of performance and interpretability in healthcare systems. Some AI models (e.g., “XGBoost (eXtreme Gradient Boosting)”) offer remarkable accuracy, but their predictions are difficult to explain to clinicians for treatment validation. The interpretability of AI models in healthcare is crucial to meet regulatory constraints (e.g., General Data Protection Regulation (GDPR), Health Insurance Portability and Accountability Act (HIPAA)), which mandate traceability of automated decisions. This paradox is compounded by the limitations of existing explainability methods (SHAP, LIME), which suffer from instability and often produce explanations that are too technical for clinical workflows. The central problem, therefore, is to design a methodology that preserves predictive effectiveness while meeting the practical needs of clinical environments. We design an ontology-based healthcare model to improve both attributes of performance and interpretability.