- Vehicle-train collisions at railroad crossings is a significant problem in North America. Railroads are earnest in their efforts to reduce the number of incidents, through safety awareness programs, and processes within their own companies.
- One railroad in particular earmarks a significant amount of money to focus on safety improvements for the engineering department to undertake. But with tens of thousands of crossings in the network, they needed to know where to focus to make the largest impact.
- The railroad had tried a retrospective approach – crossings that had experienced incidents in the past were likely to have incidents again. While there is some truth to that, in the spirit of “past performance is no guarantee of future results” we wanted to identify risky crossings that may have not yet had an incident at all.
- Borrowing from concepts from risk modeling in the insurance industry, we built models to assess the potential frequency and severity of collisions at each crossing. The models used attributes of the crossing, rather than simple past history, to quantify the risk.
- We provided a risk score for each crossing, based on a combination of the likelihood of an incident, as well as the severity of an incident, were it to occur. This allowed the railroad to rank order the crossings to begin immediately addressing through engineering remedies.
- The first two years after implementation saw a 27% and 3% year-over-year reduction in incidents, respectively.