SDP-based machine learning for flight cadets’ situational awareness

Purpose This study aims to evaluate flight cadets’ situational awareness (SA) during crosswind landing scenarios to enhance flight performance and optimize training strategies by developing a facial expression-based assessment method tailored to civil aviation training contexts. Design/methodology/approach Cadets performed crosswind landing simulations on an Intermediate Trainer Aircraft Simulator while facial videos were recorded to extract multimodal features, including Action Units (AUs) and gaze directions. Symmetric Dot Pattern (SDP) analysis quantified variations in curvature, thickness, geometric center position and regional orientation over time. Clustering algorithms were used to identify measurable patterns between situational awareness rating technique -rated SA levels and extracted features, which were then classified using Random Forest, XGBoost, LightGBM and AdaBoost models with hyperparameter tuning and key feature selection. Findings Experimental results show that combining AUs and gaze features with the optimized LightGBM algorithm can effectively support the trainees’ SA ability assessment in flight training. Practical implications Provides an objective, data-driven tool for assessing cadets’ SA in flight simulations, supporting targeted training strategies and improving aviation safety. Originality/value Among the first to integrate facial AUs, gaze analysis and SDP technology with advanced machine learning for SA assessment, offering a novel, non-intrusive approach bridging human factors and intelligent aviation training systems.