Enhancing Runway Configuration Assistant Model: The Role of Explainable AI for Model Interpretability

In the Air Traffic Management (ATM) domain, machine learning-based systems are poised to be deployed as a decision-making aid to controllers. However, the inherent complexity of these models makes it difficult to understand the machine recommendations, which is critical for acceptance in a safety-critical domain such as ATM. This research explores the use of explainable Artificial Intelligence (XAI) techniques, such as SHapley Additive exPlanations (SHAP) and permutation of features, to enhance the interpretability of the models being exercised in ATM. As our use case, we chose runway configuration assistance (RCA) [1], a specific tool that predicts the optimal runway configuration to be deployed on an airport surface. We used data from three major US airports: Charlotte Douglas International Airport (CLT), Denver International Airport (DEN), and Dallas/Fort Worth International Airport (DFW). As expected, the data shows that wind components, cloud ceiling, and visibility are the most critical factors determining the tool output. In addition, at DEN, the hour of the day became the other contributor to the decision made by the RCA tool. This work reinforces the importance of interpretability for the acceptance of AI models, specifically in safety-critical domains such as ATM.