Modeling Freeway Tunnel Driving Risk: A Driving Simulator Study Incorporating Driver Visual Characteristics

The mechanisms through which roadway design parameters and drivers’ visual characteristics influence driving risk remain unclear. This study transforms the 2-dimensional alignment indicators into 2.5-dimensional indicators perceived by drivers, and proposes a nine-zone tunnel zoning method to capture driving risk. A roadway model with eight tunnels was built for a driving-simulation experiment. Driving-behavior data were collected and converted by factor analysis into a quantitative driving-risk index. Using these perceptual indicators, zone-specific indices, alignment design parameters, and tunnel-type variables, a machine-learning-based predictor was trained to predict the risk index and examine marginal and interaction effects. Results show that the visual metrics effectively capture drivers’ visual behavior inside tunnels, and the nine-zone scheme represents risk profiles along the tunnel corridor. Factor analysis extracts longitudinal and lateral latent factors that jointly determine crash risk. The extreme gradient boosting model outperforms random forest, support vector machine, and back-propagation neural network, achieving an R 2 of 0.847, root mean square error of 1.974, and mean absolute error of 1.387. Feature-importance analysis identifies entrance section, perceptual curvature in visual zones 2 and 3, transition section, medium-length tunnel, and perceptual curved length in zone 3 as the dominant risk factors. Marginal-effect analysis indicates that entrance zones markedly increase driving risk, while transition sections and medium tunnels show weak marginal effects; higher perceptual curvature or perceptual curved length mainly increases risk. Interaction effects reveal non-linear superposition between visual and design variables, with the entrance section showing the clearest amplification effect. The study offers new insights into crash-risk mechanisms from the perspective of driver vision and helps mitigate tunnel safety problems.