Data and code underlying the online perceived risk study

This package contains data and code for a study of perceived risk during automated vehicle interactions. It includes a large scale online dataset of more than 140,000 discrete clipwise perceived risk ratings from 2,164 participants across scripted traffic scenarios, together with the corresponding simulation kinematic data. The package also provides analysis code used to infer continuous time perceived risk evolution from discrete ratings, perform validation analyses, calibrate baseline models, train and evaluate DNN prediction models, conduct SHAP attribution and cue reweighting analyses, and generate figures. The implementation includes MATLAB and Python workflows and is organised for reproducible data processing and analysis.