Transparent and Explainable AI for Liver Disease Diagnosis Using SHAP and LIME

The liver is a critical organ responsible for diverse physiological functions, yet it remains highly susceptible to damage from alcohol, medications, and metabolic imbalances. Conventional diagnostic approaches, such as blood panels and imaging, are often costly and time-consuming, delaying early detection. This study presents an interpretable machine learning framework designed to enhance liver disease prognosis by leveraging ensemble methods, including Random Forest, Gradient Boosting, XGBoost, and LightGBM. To address real-world data challenges, the framework incorporates advanced preprocessing techniques, including dimensionality optimization and class rebalancing strategies, ensuring robust model generalization. Beyond predictive performance, the model integrates state-of-the-art explainability tools—SHAP and LIME—to offer both global and local interpretability of predictions. SHAP summary, force, and dependence plots highlight feature importance and interactions, while LIME provides transparent, instance-level justifications. This dual-layered interpretability facilitates clinician trust, offering a transparent and scalable decision support system for early, explainable, and accessible liver disorder diagnosis, particularly in resource-limited healthcare settings.