Evaluating the effectiveness of safety countermeasures at highway–railway grade crossing based on a machine learning framework

The results highlight XGBoost's capability to identify important collision and inventory factors, successfully uncovering 20 of the most important factors. The DCLA clustering method forms 4 distinct groups marked by substantial internal similarity within each cluster. This approach contributes to a clearer understanding of how each countermeasure impacts collision frequency. The findings highlight the varying effectiveness of different countermeasures across clusters, improving decision making for safety at HRGCs. The study highlights the efficacy of crossbucks in addressing safety concerns during moderate traffic conditions, particularly evident in environments with a highway speed limit between 100 and 125 mph. Additionally, bells demonstrate notable effectiveness in areas with sharper track angles.