Do you think the track assignment is something that's learnable? I feel like there is some schedule where, under normal operations, the train will end up. But then there are unpredictable events that cause this to change.
I think it is. Like in the article, you can see that given certain parameters like train number and time, you predictably end up with a certain track assignment. If you can also consider track assignments leading up to this time, you can discount those tracks. For example, the 8:02PM Long Beach train always leaves on track 18, except when any 7:56 train left on track 18. As a novice to ML it's currently beyond my ability to model, but my intuition is that it should be doable.