Developing a Data-Driven Network Screening Procedure for Systemic Safety Approach

Abstract

Systemic analysis is considered an important safety analysis approach that is complementary to the Highway Safety Manual hotspot analysis. The network screening step in systemic analysis is to identify sites with characteristics that are associated with specific types of severe crashes. Traditionally, determining the risk scores involves subjective criteria. This research aims to develop a data-driven approach to replace the subjective methods used in the past. To achieve the research objective, this study collected roadway and crash data from the Connecticut Department of Transportation. A data-driven crash risk factor categorization methodology is proposed to estimate accurately the performance measures indicating crash risks. Moreover, this study proposes and compares four different risk scoring matrices to identify an optimal risk scoring method that is attuned with the principles of the systemic approach to safety as well as providing additional insights on justifying the systemic safety analysis results. The proposed methodology is implemented to conduct network screening for severe roadway departure crashes and later validated using severe aggressive-driving related crashes. Risk-based network screening results indicate that risk scores derived from normalized crash over-representation provide additional emphasis on sites with low traffic volume that are associated with high severe crash counts. The highest modified crash rate was obtained using normalized crash over-representation based risk scores indicating that the proposed network screening methodology can not only identify roadway attributes that are correlated with severe crashes but also account for low-volume roadway sites with severe crashes. The validation analysis indicated proposed method is transferrable to different emphasis area related crashes.