Toward Safer Human–AI Teaming: Phase-Specific Recognition and Interpretation of Automation Trust in Air Traffic Control

In safety-critical domains such as air traffic control (ATC), appropriate trust in AI is essential for operational effectiveness and safety. This study proposes a phase-specific framework for recognizing automation trust using behavioral and eye-tracking features across monitoring, decision-making, and whole-task phases. Data were collected through a trust-probe experiment and analyzed using linear mixed models to identify phase-specific effects of trust. Significant features were then used to train machine-learning classifiers. Results showed that trust was manifested differently across phases, with decision-making producing the most informative eye-tracking indicators, including fixation duration and saccade velocity. Models trained on multi-phase data outperformed single-phase models, and Gradient Boosting achieved the highest accuracy of 86.76%. SHapley Additive exPlanations were further used to identify influential features and their directional effects. The framework supports transparent trust recognition and informs human-centered automation, safety management, and policy development in high-stakes environments.