Network-level crash risk analysis using large-scale geometry features.

Road traffic crashes are common occurrences that create substantial losses and hazards to society. A complex interaction of components, including drivers, vehicles, roads, and the environment, can impact the causes of these crashes. Due to its complexity, crash identification, and prediction research over large-scale areas faces several obstacles, including high costs and challenging data collecting. This study offers a method for large-scale road network crash risk identification based on open-source data, given that roadways' horizontal and vertical geometric alignment is crucial in highway traffic crashes. This methodology includes a comprehensive technique for feature extraction from horizontal curves (H-curves) and vertical curves (V-curves) and a novel way of combining the XGBoost model's attributes with the Harris Hawks Optimization (HHO) algorithm-referred to as the HHO-XGBoost model. Using this model on the road geometry-crash risk dataset developed specifically for this study, the HHO approach adaptively identifies the optimal set of XGBoost hyperparameters and yields favorable outcomes. This study creates a three-dimensional road geometry database that may be utilized for various road infrastructure management, operation, and safety in addition to completing a tiered risk analysis of "region-road-segment" for large-scale road networks. It also offers direction on using swarm intelligence algorithms in integrated learning models.