Interpretable AI-Enhanced Multi-Criteria Decision-Making for Time-Critical Emergency Applications With a Practical Application to Dynamic Alternate Airport Selection in Aviation

Dies ist die aktuellste Version dieses Eintrags.

Kurzfassung

Time-critical decision making in emergency situations poses significant challenges for both humans and traditional multi-criteria decision-making methods. Under such demanding conditions, supporting human decision makers with tools that structure information and reduce cognitive burden is both beneficial and prudent. This thesis addresses this challenge by introducing the ExACT-MCDM methodology, a framework that integrates artificial intelligence with multi-criteria decision-making concepts to support high-stakes decisions in environments where human operators remain responsible and actively involved in the emergency scenario.

The proposed methodology reformulates emergency, time-critical multi-criteria decision-making problems as supervised ranking tasks, enabling the use of machine learning models that preserve the structure of options and attributes while providing rapid, context-aware recommendations. To address the scarcity of real emergency data, a synthetic scenario generation and labeling process was developed, incorporating expert knowledge through structured scoring functions with hard and soft constraints. Robustness was achieved by simulating realistic variability in contextual factors and maintaining their interdependencies, resulting in datasets that reflect the uncertainty inherent in real emergency operations.

The methodology was applied to the dynamic alternate-airport selection problem, a representative aviation emergency scenario in which pilots must select a diversion airport under dynamic and unfamiliar conditions. Experiments were conducted using both classification and ranking formulations across a range of models. The results indicated that ranking was better suited to the structure of the problem and more consistent with real decision-making processes, where the goal is to order feasible alternatives rather than output a single preferred choice. Interpretability and explainability were incorporated throughout the process using inherently interpretable models, structured feature contributions, and model-agnostic tools such as SHAP, with the aim of helping stakeholders maintain oversight of why recommendations are made.

The results demonstrate the potential that artificial intelligence has to enhance multi-criteria decision making and provide fast, context-sensitive, and transparent decision support suitable for emergency situations while preserving the responsibility and authority of human decision makers. Aside from the ExACT-MCDM methodology, this thesis contributes a scalable benchmark that advances understanding of the dynamic alternate-airport selection problem and provides guidance for integrating such models into operational systems. Together, these contributions demonstrate how artificial intelligence can meaningfully augment decision making in complex, unpredictable, and high-stakes environments, including cockpit decision-support tools such as the Intelligent Pilot Advisory System under development at the German Aerospace Center.

Verfügbare Versionen dieses Eintrags

• Interpretable AI-Enhanced Multi-Criteria Decision-Making for Time-Critical Emergency Applications With a Practical Application to Dynamic Alternate Airport Selection in Aviation. (deposited 10 Jun 2026 10:29) [Gegenwärtig angezeigt]