Fostering trust and interpretability: integrating explainable AI (XAI) with machine learning for enhanced disease prediction and decision transparency

Abstract

Medical healthcare has advanced substantially due to advancements in Artificial Intelligence (AI) techniques for early disease detection alongside support for clinical decisions. However, a gap exists in widespread adoption of results of these algorithms by public due to black box nature of models. The undisclosed nature of these systems creates fundamental obstacles within medical sectors that handle crucial cases because medical practitioners needs to understand the reasoning behind the outcome of a particular disease. A hybrid Machine Learning (ML) framework integrating Explainable AI (XAI) strategies that will improve both predictive performance and interpretability is explored in proposed work. The system leverages Decision Trees, Naive Bayes, Random Forests and XGBoost algorithms to predict the medical condition risks of Diabetes, Anaemia, Thalassemia, Heart Disease, Thrombocytopenia within its framework. SHAP (SHapley Additive exPlanations) together with LIME (Local Interpretable Model-agnostic Explanations) adds functionality to the proposed system by displaying important features contributing to each prediction. The framework upholds an accuracy of 99.2% besides the ability to provide understandable explanations for interpretation of model outputs. The performance combined with interpretability from the framework enables clinical practitioners to make decisions through an understanding of AI-generated outputs thereby reducing distrust in AI-driven healthcare. The proposed manuscript suggests the development of a hybrid Model with machine learning Techniques and includes interpretability behind outcomes: ML-XAI framework for predicting diseases. Five Diseases are considered in the proposed work: Diabetes, Anaemia, Thalassemia, Heart Disease, and Thrombocytopenia. Black Box nature of Most Machine Learning models leads to distrust and acceptance about results especially in medical domain. The proposed model solves this issue by diagnosing disease with ML models and providing attributes which are responsible for disease being diagnosed by models.

The authors' abstract, as published at the source. Diagnostic Pathology, 2025 · DOI ↗