We have reached now a computation power where machine learning algorithms can easily be applied across a spectrum of industries to resolve a huge variety of problems. While first successes are also published in the railway industry this also opens up the market for what I call “snake-oil-resellers”. Companies or individuals that use publicly available models, claim that they have deeper understanding of the topic and find companies that give them large amounts of money for a bad solution or for buying into a lock-in situation. While the obvious solution is to build up own competencies, it is not always that easy because of time and investment. This is why I put here some recommendations how railway or transportation companies can approach the topic.

Suggested approaches

• Correlations does not substitute understanding of causality
  • We unfortunately confuse these two concepts most often since we believe that a strong correlation must have some causality effects. This is why technical technical experts should collaborate with the data scientists to discuss where it could be causility or where it is rather a correlation.
• Build up a culture of continuous development
  • Early results will most likely not be very promising. But going further down the path and also accepting failures, the likelyhood of success will increase. There is no single/obvious path to an optimal solution and people telling you this do not have good intentions. A culture of trial and error with continious development is necessary to also find out when perceived causailites turn out to be simple correlations.
• Having a good training set – but also questioning that again and again
  • The quality of the output reflects the quality of the input. If the data available is of questionable quality, even the best algos will not create a good solution.
• Avoid lock-ins
  • Do never invest in a system where you become dependable on the supplier. If it is not successful, any investment is lost. If it is successful you will pay heavily over the next years.
• Question and discuss with a solution provider
  • Any supplier should be a partner that points out strength and shortcomings of the suggested approaches. Open discussions and questions of the “black-box” are necessary to build on both sides sufficient trust.
• Solve problems that are worth solving
  • Focus on topics that have an immediate value regarding availability or costs.

When following these recommendations, I believe it will be possible to create value from algos in the railway sector. But we should not overestimate it and see it as the solution for every possible problem – it is one of many tools of the RAMS people but should definitely not remain the only one.