Get started
Work

A Hybrid Explainable AI Framework for Fair Automated Decision-Making in Healthcare

Year: 2026
Type: conference-paper
Abstract: Artificial Intelligence (AI) is increasingly used in healthcare to support high-stakes Automated Decision-Making (ADM), such as disease detection and risk stratification. While these systems can achieve high predictive performance, their opaque “black box” behaviour raises serious concerns about interpretability, fairness, and accountability. This paper proposes a hybrid Explainable AI framework that jointly addresses model transparency and subgroup fairness in a single workflow. Using the NIH Chest X-ray14 dataset, we constructed a dual pipeline to process a structured metadata and chest radiographs for binary classification into Effusion and No Effusion. Models based on Logistic Regression, Random Forest, and Boost were fitted to the tabular data, while a Convolutional Neural Network was trained on pre-processed radiographs. Class imbalance was managed through class weights, oversampling, and SMOTE. Explainability was provided using SHAP for global and local feature importance, LIME for instance-level explanations, and Grad-CAM for spatial visualisation of CNN predictions. Fairness was quantified using Demographic Parity Difference and Equalized Odds Difference across gender and age groups, with mitigation using Exponentiated Gradient and Threshold Optimizer methods. Results show that balancing strategies improve Effusion recall in both pipelines, while hybrid explanations offer complementary insights that align with clinical reasoning. Fairness analysis reveals measurable subgroup disparities that can be reduced without severely degrading performance. The proposed framework demonstrates how interpretability and fairness can be integrated into healthcare ADM, providing a reproducible pathway towards more trustworthy and accountable AI systems. Keywords Artificial Intelligence (AI), Automated Decision-Making (ADM), Explainable AI (XAI), Convolutional Neural Network (CNN), Fairness, Healthcare
Language: English
Open Access status: green
OpenAlex ID: W7134061160
FWCI:
Cites:
Cited by:
Related to: