Regional Corridor-Level Approach to Predict Wrong-Way Driving Crashes and Identify Wrong-Way Driving Hotspots on Florida Arterials

Abstract

Previous research on wrong-way driving crashes (WWCs) has largely focused on limited access roadways, although arterial roads are more prone to these crashes. Since it is often unknown exactly where arterial WWCs (AWWCs) originate, studying AWWCs on corridors can be more effective than studying individual segments or intersections. This paper introduces an innovative regional corridor-level approach to identify AWWC attributes and hotspots in South Florida. Three distinct corridor definitions were developed based on area type and either speed limit, annual average daily traffic (AADT), or lane count. For each definition, annual AWWC frequency was predicted using regression and machine learning models. Predictor variables included macroscopic corridor features and non-crash WWD events (which have not previously been used to predict AWWCs). A negative binomial model with interaction terms was the best-fitting model for each definition. Comparing the definitions showed that using area type and AADT to define corridors resulted in the most accurate and informative model, with urban corridors and corridors with higher WWD citation counts likely to experience more AWWCs. Speed limit, intersection density, and AADT were positively associated with AWWCs at low levels of these variables and negatively associated with AWWCs at high levels of these variables. Ten hotspot corridors were chosen based on their high predicted and observed AWWC counts. All these hotspots were in urban areas and had high speed limits and AADT, with multiple hotspots found along US 1. This macroscopic corridor-level approach can be transferred to other regions and states to quickly identify AWWC hotspots.