Helping ProRail optimize its operations with predictive models

Executive Summary

ProRail, the appointed railway infrastructure manager in the Netherlands, has partnered with Xomnia to develop two predictive models: The first to help ProRail in forecasting the wearing out of different parts of the rails on the long-term, and the second to help ProRail’s inspectors determine with more accuracy the times and locations that have a higher likelihood of trespassers going over the tracks.

ProRail collaborated with Xomnia and other partners to develop the first model, which uses machine learning to estimate what parts of the rail will be worn out within the 5-10 years. The model has been put in production by the railway manager for two years.

To address the trespasser challenge, Xomnia helped ProRail develop a dashboard that uses machine learning to predict the times and locations with a higher risk of people trespassing along the train tracks. The dashboard has the potential to help ProRail focus its surveillance capacity more effectively.

"Xomnia has helped us with setting up our DataLab, by providing junior data scientists and infrastructure expertise.
Paul van der Voort, Program manager DataLab


As the manager of railway infrastructure across the Netherlands, ProRail is responsible for determining and planning which parts of the railway need to be replaced due to being worn out. By having those estimations ready on a 5-10 year range, ProRail can ensure having the needed budgets and maintenance plans ready on time.

Previously, maintenance specialists relied on industry standards or rules of thumb to estimate when different parts of the rail will be worn out and need to be replaced. This approach, however, couldn’t take into account all the variables that affect the lifespan of a track, such as the different types of trains travelling at different speeds, etc. This resulted sometimes in inaccurate estimations that led to replacing parts of the rail that are not really worn out yet, and which could have still been used for several years.

A second challenge that ProRail sought to mitigate with machine learning was trespasser control. Trespassers are among the top 3 causes of train delays that are directly related to the railway. Everyday, train drivers see a number of trespassers, leading them to call traffic control to report the trespassers, and causing them and several other trains along the tracks to slow down. This consequently results in train delays.


To address the first challenge, ProRail reached out to Xomnia and other partners to develop a machine learning model that predicts when a certain piece of the rail needs replacement. The model measures for each part of the rails the wear over many years, and uses regression to extrapolate this to a predicted wear in 5-10 years . The predictions from the model are used as a starting point by the maintenance experts at ProRail, who then make replacement plans.

To address the trespasser challenge, ProRail collaborated with a junior data scientist from Xomnia for about a year to develop a proof of concept for a trespasser detection machine learning model, particularly utilizing support-vector machines. The model was given different kinds of data linked to an increased likelihood of trespassing, like where a trespasser has been seen in the past by train drivers, as well as data about external influences such as the weather, the proximity to schools or sports events, holidays, and others.

The model calculated per 200m of the track the likelihood of a trespasser being on the tracks, based on the previously mentioned factors, and gave a warning of higher chances of trespassers. Using this model, railway inspectors can go to the places that have the highest risk at the times of highest likelihood, increasing the chance of seeing a trespasser by multiple times.  


The trespassing model has proven to be a successful pilot that is capable of improving ProRail’s processes. The predictive model of railway wear has been put in production since 2019, and ProRail has kept it up to date by adding new parameters to it to reflect changes in the numbers of trains, train schedules and types and speeds of trains. Taking all those new parameters into account, the model is more complex and provides different inputs, helping ProRail make more accurate predictions and saving maintenance costs.