Explainable AI Models for Disease Diagnosis and Prediction in Healthcare Decision Support Systems

Abstract

The rapid adoption of Artificial Intelligence (AI) in healthcare has highlighted the need for predictive systems that are not only accurate but also interpretable and transparent. This work presents the development of Explainable AI (XAI)-driven Healthcare Decision Support Systems (HDSS) for disease diagnosis and prediction using structured and unstructured datasets. Four datasets are employed: PIMA Diabetes (structured clinical data), Self-reported Mental Health Diagnoses (SMHD), Dreaddit, and GoEmotions (mental health and emotion detection from social media). The methodology is involved data preparation, Chi-Square feature selection, and model building using Generalized Additive Models (GAMs) and Explainable Boosting Machines (EBMs). To enhance interpretability, K-LIME is applied for subgroup-level explanations and Fast SHAP for patient- and text-level attributions. Results are demonstrated that EBMs consistently outperformed GAMs in predictive accuracy while retaining interpretability (AUC range: 0.81–0.89). K-LIME successfully revealed meaningful patient/stress/emotion clusters, whereas Fast SHAP provided real-time, individualized explanations. The work concludes that XAI-based model is achieved near state-of-the-art predictive performance while ensuring transparency, scalability, and clinical trustworthiness, making them suitable for real-world healthcare decision support.