Explainable Artificial Intelligence in High-Stakes Decision Systems: A Systematic Review (2015–2025)

Zeeshan Ashraf

Background and Objective: Artificial Intelligence systems are increasingly deployed in high-stakes domains, including healthcare, finance, and legal systems. The opacity of complex machine learning models raises concerns regarding accountability, fairness, and regulatory compliance. To systematically review Explainable Artificial Intelligence (XAI) methodologies, evaluation standards, and regulatory implications in high-risk decision systems from 2015 to 2025. Materials and Methods: A PRISMA-guided systematic review was conducted across Scopus, IEEE Xplore, ACM Digital Library, and Web of Science. Peer-reviewed journal articles published in English between January 2015 and January 2025 were included. Results: A total of 118 studies met the inclusion criteria. Healthcare accounted for 48% (n = 57) of studies, finance 32% (n = 38), legal systems 15% (n = 18), and other public-sector domains 5% (n = 5). A post-hoc explanation methods dominated (61%, n = 72), while intrinsic interpretable models comprised 39% (n = 46). Fidelity metrics were used in 74% (n = 87) of studies; however, only 27% (n = 32) conducted human-centered validation. Conclusion: Although, XAI research is expanding rapidly, standardized evaluation metrics and harmonized regulatory integration remain limited. Human-centered validation and domain-specific benchmarks are essential for responsible AI deployment in high-stakes systems.

View Fulltext Back