Using Spatial Tools to Analyze Crash and Roadway Data, Project Number 0605(08)

The Pennsylvania Department of Transportation (PennDOT) engaged Gannett Fleming to conduct research into best practices in the use of geospatial analysis tools for highway safety analyses. The goals of the effort were to define a methodology for PennDOT to follow in identifying the best candidate locations for highway safety improvements, and to develop a Proof of Concept to test the proposed methodology. After conducting interviews and workshops involving more than 35 of PennDOT's stakeholders in highway safety processes, Gannett Fleming interviewed highway safety managers in five other state and federal highway agencies to determine what innovative tools and practices are currently being used. Gannett Fleming's research also included a review of literature related to the study from more than 80 sources. Based on Gannett Fleming's research and analysis, PennDOT selected the Highway Safety Data Relationships Knowledge Base for further research. The knowledge base is an information repository based on concepts in data mining and expert systems. It uses advanced statistical analysis methods and expert business knowledge rules to discover data patterns based on correlation and other forms of relationships in the data. The knowledge base can be applied to diagnosing specific combinations of data attributes and features that may indicate the causative factors among homogeneous populations of crashes. Most highway safety data analyses involve studying correlations among multiple data sets. The knowledge base is an innovative and comprehensive tool for such an application. It provides a framework for identifying and managing relationships among many combinations of data sets that are useful in highway safety analyses. Gannett Fleming proceeded to develop a prototype as a proof of concept. Gannett Fleming demonstrated the prototype using actual PennDOT crash data. Three analysis scenarios were demonstrated: evaluating safety programming alternatives for alcohol-involved crashes, diagnosing data patterns of crashes at a selected highway location, and identifying potential sites for system-wide deployment of a selected countermeasure.