Skip to content

The Opportunity

A Class I railroad began focusing on reducing human factor derailments.  Though incidents reported to the Federal Railroad Administration had dropped in the previous twelve months, the safety department was striving for further improvement.

They wanted to use data and analytics to help drive more declines in reportable derailments that were attributable to employees.

Our Approach

We established a safety data repository drawing from a client data lake.  These data contained a variety of potentially related factors.  This allowed us to go back and reconstruct the facts and context for incidents for when and employee was involved in a derailment or rule violation.

A statistical risk model produced risk scores for employees, showing the likelihood of them being involved in a derailment or rules violation incident in the next 30 days

Factors included discipline events, work schedule, rules exams, attendance and absenteeism, drug and alcohol tests, furlough, and human resource information.  We also performed text mining on incident reports to augment this structured data with themes..

The employee-level scores and quantification of the risk factors were built into an interactive PowerBI application.

The Impact

A holdout analysis of the model showed that for the top 5% riskiest employees, the model could correctly capture incidents 3.1 times better than just randomly guessing.  This made better use of manager time by allowing them to focus on the top riskiest employees.

Though not fully attributable to the risk model, the application was significant contributor to a 21% decline in reportables in the following two years.

Back To Top