Towards Transparent, Fair, and Trustworthy AI for Sustainable Decisions in High-Stakes Sectors: Research Methodology and Experimental Design

This paper presents the research methodology and experimental design for a Postdoc research program investigating transparent, fair, and trustworthy AI across healthcare, finance, and environmental sustainability. Our first conference literature review identified five critical gaps in the existing research. These gaps led to four hypotheses (H1–H4), which we tested through five pre-registered experiments. The experiments used three benchmark datasets: MIMIC-III for healthcare, Home Credit Default Risk of finance, and UCI Air Quality for environmental research. The Trustworthy AI Assessment Composite (TAAC = α·Ī + β·F̄ + γ·R̄) combines normalized scores for interpretability, fairness, and robustness into a validated composite measure. The evaluation used SHAP, LIME, integrated gradients, adversarial debiasing, and perturbation-based robustness testing and was validated by 24 domain experts. A rigorous statistical plan, including Bonferroni correction and bootstrap confidence intervals, governs all tests.