Using Eye-Tracking Data to Predict Situation Awareness in Real Time During Takeover Transitions in Conditionally Automated Driving

Situation awareness (SA) is critical to improving takeover performance during the transition period from automated driving manual driving. Although many studies measured SA or after task, few have attempted predict in real time In this work, we propose conditionally using eye-tracking and self-reported data. First, a tree ensemble machine learning model, named LightGBM (Light Gradient Boosting Machine), was used SA. Second, order understand what factors influenced how, SHAP (SHapley Additive exPlanations) values of individual predictor variables model were calculated. These explained prediction by identifying most important their effects on SA, which further improved through feature selection. We standardized between 0 1 aggregating three measures (i.e., placement, distance, speed estimation vehicles with regard ego-vehicle) recreating simulated scenarios, 33 participants viewed 32 videos six lengths 20 s. Using only data, our proposed outperformed other selected models, having root-mean-squared error (RMSE) 0.121, mean absolute (MAE) 0.096, 0.719 correlation coefficient predicted ground truth. The code available at https://github.com/refengchou/Situation-awareness-prediction. Our provided implications how monitor