Explainable AI in Healthcare Analytics and Clinical Decision Support Systems: Methods and Experiments
Contributors
Dr. Prakash Mohan
Prof. (Dr.) Shashi Kant Gupta
Keywords
Proceeding
Track
General Track
License
Copyright (c) 2026 Sustainable Global Societies Initiative

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Abstract
AI tools can help with clinical risk assessment, but their predictions are opaque and results are unpredictable; this restricts the use of AI in clinical decision support systems (CDSS). This study proposes a methodology that is repeatable for the combination of mortality prediction, explainability, and fairness auditing. A literature review was conducted first, and then a MIMIC-IV admission-level demonstration was developed, which includes 100 patients and 275 admissions with 15 mortality outcomes. Data preparation includes leakage-variable exclusion, imputation, encoding and model development with logistic regression, random forest and gradient boosting. For the full evaluation, a patient-grouped validation protocol, SHAP/LIME explanations, discrimination and calibration metrics as well as subgroup checks are proposed. The second-conference presentation reports a preliminary logistic-regression AUROC of 0.732, which is reported as the only experimental result, without claiming that the suggested explanation and fairness assessment have been validated. The design offers an explicit ground for further assessment of clinically meaningful predictions, and emphasizes that the small, highly skewed demonstration cohort is a limitation.