Imprecise Probability Theory and Decision Making

Imprecise probability theory extends classical probability by allowing uncertainty to be represented through sets of probability measures rather than single distributions. This approach acknowledges epistemic imprecision arising from limited data, expert disagreement or inherent variability. By employing lower and upper probability bounds, credal sets and related constructs, decision makers can capture both aleatory and epistemic uncertainty in a coherent framework. Decision rules under imprecise probabilities encompass admissibility, maximality, Γ-maximin, E-admissibility and other criteria that generalise expected utility. Such rules offer robust choices when precise probabilities are unavailable or when sensitivity to probability estimates is intolerable. Computationally, challenges arise in evaluating decision criteria over infinite or high-dimensional credal sets, motivating the development of tractable approximations, linear and quadratic programming techniques, and specialised inference algorithms. Applications span human reliability analysis, system safety assessment, finance, machine learning and artificial intelligence, where robust decision making under severe uncertainty is essential. Recent advances have linked imprecise probability models with Choquet integrals, distortion functions and credal networks, broadening their practical reach while preserving rigorous coherence conditions.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Researchers have proposed distortion-model approaches for estimating human error probabilities in safety-critical systems. By integrating imprecise probabilities with Bayesian network structures and distortion functions, these models yield robust error-rate estimates that remain stable under small perturbations of input data. Comparative analyses demonstrate that distortion-based estimators align with previous techniques while improving resilience to missing or conflicting information.

In operations research, novel methods for inner approximations of coherent lower probabilities have been developed to streamline decision making under severe uncertainty. By transforming general credal sets into subclasses with desirable mathematical properties—such as supermodularity or complete monotonicity—these methods recast decision problems as linear or quadratic programmes. The resulting approximations are at least as informative as the originals, facilitating efficient computation of optimal alternatives and illustrating significant gains in practice.

A subjective interpretation of credibility measures and expectations has been advanced through an extension of de Finetti’s coherence principles. By framing credibility via fair-betting schemes and penalty criteria, this work unites notions of consonance, partial resolution of uncertainty and pessimism–optimism indifference. Choquet integrals underlie the expectation operators, providing a unified behavioural foundation for credal models and reinforcing the links between belief functions and imprecise probability theory.

Imprecise Probability Theory and Decision Making publication trend

The graph below shows the total number of articles in imprecise probability theory and decision making across all publications each year (not limited to Nature Index journals).

Technical terms

Imprecise probability: A representation of uncertainty by a set of probability measures or bounds, reflecting incomplete knowledge.

Credal set: A convex set of probability distributions representing all plausible beliefs about a random phenomenon.

Coherence: A behavioural consistency criterion ensuring that probability assessments avoid sure-loss in betting frameworks.

Choquet integral: A generalised integral with respect to nonadditive measures, used to compute expectations under imprecise probabilities and belief functions.

Credal network: A graphical model generalising Bayesian networks, whose local uncertainties are described by credal sets rather than single conditional probabilities.

References

1. A subjective interpretation of Liu–Liu’s credibility measures and expectations. Fuzzy Optimization and Decision Making (2023).
2. Distortion models for estimating human error probabilities. Safety Science (2023).
3. Inner approximations of coherent lower probabilities and their application to decision making problems. Annals of Operations Research (2023).