Learning Paradigms for AI-Driven Decision Support in Safety-Critical and Resource-Constrained Environments

Abstract

Artificial intelligence has become a foundational capability for decision support across domains where failures may result in safety, economic, or societal harm. These environments are often constrained by limited data, restricted computational resources, latency requirements, and evolving operational conditions. This article examines learning paradigms that underpin AI-driven decision support systems in such contexts, with emphasis on safety-critical and resource-constrained settings. By synthesizing evidence across healthcare, transportation, cybersecurity, industrial systems, and edge-enabled infrastructures, the study analyzes how supervised, unsupervised, semi-supervised, reinforcement, federated, and ensemble learning paradigms contribute to reliable decision making. A unified methodological framework is proposed, integrating architectural design, learning selection, and evaluation strategies. Empirical results and comparative analyses demonstrate trade-offs among accuracy, robustness, interpretability, and resource efficiency, highlighting pathways toward resilient and adaptive AI decision support.