A Data-Driven Multi-Perspective Solution for Improving Transit Bus Safety

This research proposes a data-driven and multi-perspective solution to reduce injuries and fatalities due to bus collisions with other vehicles. It goes beyond reviewing historical bus crash reports by collecting field LiDAR data for better understanding a variety of factors that may contribute to bus collisions and injuries/fatalities. It also recognizes that collision avoidance technologies (which in some cases may cause distractions and cognitive overload to drivers) are not the only solution to bus safety. Bus safety depends on many factors such as driver, vehicle design, roadway design, and traffic environment and requires a comprehensive set of safety strategies that complement each other. This research proposes to derive insights from both crash reports and LiDAR data to help transit agencies develop comprehensive safety improvement strategies. Specifically, it proposes to use interpretable machine learning and natural language process (NLP) to extract critical factors from crash reports and quantify their individual and joint impacts on injury severity. It also uses LiDAR to capture near-crash and other safety related events in a naturalistic setting and proposes advanced algorithms to derive insights from the captured data, building a solid data foundation for developing and evaluating various safety improvement strategies from multiple perspectives.