It may be possible to do some machine learning on the live train running API (which reports the real-time position of trains). Use the published line statuses t

It may be possible to do some machine learning on the live train running API (which reports the real-time position of trains). Use the published line statuses to teach it what normal looks like then predict suspicious service. You could crowd source data too but internet coverage is intermittent on the tube. This would work better for main line trains. There are are similar APIs for them which I've built a wrapper for: https://github.com/jpsingleton/Huxley https://github.com/jpsingleton/Huxley