Can you predict the delays?

[christophM]

Objective: Predict delays as accurately as possible.

Difference to #1 is that you ignore the interpretability of the algorithm and just aim for accuracy.

This could be helpful in planning your daily trips.

[mauriciobosshard]

I might be interested in working on this. There are a lot of variables in the delays dataset: time of day, day of week, previous delays on the same line, vehicle id, etc. Other sources might include weather, construction sites and accidents. Some of those are already being looked at for other issues.

As you pointed out during the preparation, we should think about how to limit our scope. Do we only look at the most recent data? Only at certain lines?

One final thing to consider is how to select and tune our models. Some variant of cross-validation that takes into account the chronological order of the data, like "forward chaining" might be good.

[krlmlr]

https://github.com/krlmlr/ODDPredictDelays