Quantifying Uncertainty in Probabilistic Data Analytics Models for Decision Support under Risk and Ambiguity

Abstract

Decision-making in data-intensive environments often involves varying degrees of uncertainty, particularly when risk and ambiguity are both present. Probabilistic data analytics models, such as Bayesian frameworks, support vector machines, and fuzzy systems, offer mechanisms for quantifying uncertainty. However, these models must be critically evaluated for their capacity to handle not only stochastic variability (risk) but also epistemic uncertainty (ambiguity). This paper explores key methodologies for uncertainty quantification, identifies limitations in current practices, and proposes integrative techniques that enhance robustness in decision support systems. Our findings suggest that combining probabilistic and non-probabilistic approaches (e.g., fuzzy logic, belief functions) can improve inference under deep uncertainty.