Explainable Artificial Intelligence and Predictive Analytics: Applications Across Healthcare, Finance, Cybersecurity, and Sustainable Systems
TL;DR
XAI should be treated not as a visualization added after modeling, but as a socio-technical control layer that connects predictive performance with accountability, safety, and sustainable adoption across complex, regulated, and resource-constrained operational environments.
Abstract
Artificial intelligence increasingly shapes consequential decisions, yet predictive accuracy alone is insufficient when users cannot understand, contest, or safely act on model outputs. This systematic review examines how explainable artificial intelligence (XAI) is combined with predictive analytics across healthcare, finance, cybersecurity, and sustainable systems. Following a PRISMA-informed protocol, 35 peer-reviewed studies published between 2018 and 2025 were selected from multidisciplinary databases and coded for application, model class, explanation technique, validation strategy, stakeholder, and implementation maturity. The synthesis shows that SHapley Additive exPlanations, feature importance, Local Interpretable Model-agnostic Explanations, saliency methods, and rule-based surrogates dominate current practice. Healthcare studies emphasize diagnostic reasoning, risk stratification, antimicrobial resistance, medical imaging, and supply-chain resilience; finance studies prioritize credit assessment, fraud detection, information security, and technology adoption; cybersecurity studies focus on intrusion detection, malware analysis, and analyst-centered threat interpretation; sustainable-systems research applies XAI to renewable-energy forecasting, load prediction, maintenance, and carbon-reduction planning. Across domains, explanations are most useful when they are stakeholder-specific, clinically or operationally plausible, stable under perturbation, and linked to an actionable decision. However, many studies validate predictive performance more thoroughly than explanation quality, rely on post-hoc tools without testing fidelity, and provide limited evidence from real-world users. The review proposes a cross-domain governance framework integrating model selection, explanation design, human evaluation, fairness testing, drift monitoring, and documentation. XAI should therefore be treated not as a visualization added after modeling, but as a socio-technical control layer that connects predictive performance with accountability, safety, and sustainable adoption across complex, regulated, and resource-constrained operational environments.