An NDSU researcher recently published a study in the journal Accident Analysis and Prevention, where he tested machine learning techniques to identify key causes of railroad accidents.

In the article, “Railroad Accident Analysis Using Extreme Gradient Boosting,” Raj Bridgelall, assistant professor of transportation, logistics and finance, tested 11 different types of machine learning methods and found that a method known as extreme gradient boosting was most effective at predicting accident type. The article is co-written by Denver Tolliver, NDSU Upper Great Plains Transportation Institute director.

The method showed that derailments were most closely associated with lower track classes, non-signalized areas and rail movement authorizations within areas with restricted weight and speed limits.

According to Bridgelall, a researcher with the institute, railroads lose hundreds of millions of dollars from accidents each year with derailments, accounting for more than 70 percent of the U.S. rail industry’s average annual accident cost.

Bridgelall said computer and statistical modeling and machine learning techniques that improve knowledge of factors most closely related to derailments can help railroads develop more cost-effective and impactful strategies for reducing derailments.

As a student-focused, land-grant, research university, we serve our citizens.