Application of Dynamic Crash Prediction Methodologies to FDOT Safety and Transportation System Management & Operation (TSM&O) Programs

US Transportation Collection2020PDFTech ReportWang, ZhenyuLin, Pei-SungYang, RunanVasili, AbhijitHadi, MohammedWang, TaoEcheverri, CatalinaUniversity of South Florida. Center for Urban Transportation ResearchUniversity of South FloridaFlorida International UniversityGannett Fleming, Inc.Florida. Department of TransportationFlorida. Department of TransportationFloridaUnited StatesArterial highwaysCountermeasuresCrash risk forecastingData miningFreewaysMachine learningPredictive modelsSafety managementTraffic crashesDataDynamic crash predictionproactive safety managementarterialsbig datacrash preventionBE548BE54805/2018\u201311/2020Dynamic crash prediction, a proactive safety management strategy, predicts crash risk based on prevailing traffic conditions and prevents crashes before occurrence. As an innovative technology, dynamic crash prediction provides a potential way for the Florida Department of Transportation (FDOT) to take advantage of information provided by intelligent transportation system (ITS) devices and other sources, combined with increasingly available big data/data analytics to effectively the safety and mobility of Florida roadway systems. This project documents and evaluates existing dynamic crash prediction methods and practices related to accuracy and timeliness and develops recommendations for implementing a proactive safety strategy in Florida. Information related to dynamic crash prediction, including previous studies, existing vendor and technologies, and current user and implementation experiences, was collected and assembled through literature review, online search, document review, and interviews. A comparison of existing dynamic crash prediction technologies was performed to identify the \u201cbest\u201d technology (WayCare) for the pilot study. Historical traffic and crash data (2015-2019) were collected at two study sites, covering freeway segments (I-95) and an arterial corridor (E Sunrise Blvd) in FDOT District 4. Based on the historical data, WayCare built a prediction model that can predict crash risk for a three-hour time window based on traffic and crash conditions for nine hours prior to the time window. With the WayCare model, the research team conducted an offline test, using three-month data in 2020 (January, February, July), to evaluate the performance of dynamic crash prediction in the Florida roadway environment. The offline results showed that the WayCare model presented better performance for the I-95 site than the E Sunrise Blvd site due to the high number of crashes and relative simplicity of traffic patterns on freeways. The model can correctly predict 60% crashes during the PM period (3:00\u20136:00 PM) on I-95. Based on pilot study results, it is suggested to implement the dynamic prediction model preferentially on freeways but work with the WayCare team to improve performance of its model for periods other than the PM period. Three crash prediction actions (DMS safety messages, stationary police cars with flashing lights, and advance warning to Road Rangers) were proposed based on WayCare\u2019s experience and the availability of TSM&O applications in FDOT District 4. A further study is needed to address the safety and mobility of crash prediction actions and real-time data connection.125