Improving decision making under uncertainty with data analytics: Bayesian networks, reinforcement learning, and risk perception feedback for disaster management

Decision-making under uncertainty is one of the most critical challenges in disaster management, where human risk perception is the most important factor that shapes responses to crises. This paper introduces a hybrid risk-based decision model that integrates behavioral science, probabilistic risk assessment, and machine learning in an analysis of how individuals and organizations make high-stakes decisions under uncertainty. Our model fuses cognitive biases with dynamic risk evaluation to capture real-time fluctuations in decision-making behavior during disasters. While conventional models rely on the basis of rational decision-making, we introduce bounded rationality, emotional triggers, and social influence factors to make it more realistic. Bayesian networks and agent-based simulations are applied in the proposed framework to forecast adaptive decision pathways and gain insight into the evolution of uncertainty perception over time. The model is tested by case studies on natural disasters, pandemics, and industrial accidents and uncovers the key patterns of delays in response, overconfidence, and risk-averse attitudes. Extensive results show how decision inertia, misinformation, and cognitive overload under stress violate many axioms of traditional risk models. We propose a new "risk-perception feedback loop" by integrating real-time information with AI-driven adaptive decision-making tools, aiming at enhanced resilience within disaster scenarios. The study blends computational modelling with psychological insight, yielding a revolutionary approach to both understanding and better decision-making in the face of extreme uncertainty, with a strong possibility of profoundly smarter disaster response systems and significantly more effective risk communication strategies.