You know, the funny thing about flightcaster. I was in the airport and my flight was cancelled. 2 hours later, I checked flightcaster and it said the flight was on time!

If you can't tell when a flight really is cancelled 2 hours after the fact, there's something wrong with the algorithm.

Wow. I'm sorry to hear that. That is simply unacceptable. We pride ourselves on great data, but we do make mistakes (bugs, data anomolies, etc).

If you decide to give us another shot, please let us know if there's ever an issue--we'll troubleshoot it immediately. Our best feedback comes from HN readers.

We're working on ways to help report when a prediction is off. Thanks for using our product. If there's anything I personally can do to help, please dm me at @flightcaster or email jason@flightcaster.com

I thought it would be a good chance to test out the accuracy. The airline's website had been updated, so I know the information was "out there" about the flight status. I know you are using predictive algorithms, but you may also try to get some kind of feed for actual flight status when possible. I know that kind of polling is expensive, but it's probably the only way to avoid this problem in the future.

This particular flight was canceled because something was wrong with the plane. There were guys working on it for an hour before the flight was due to leave.

Thanks for the response Jason. It's a good service, this may have been an anomaly.

It seems like in this situation Flightcaster might benefit from doing some sort of sentiment analysis on twitter and cross check that with their machine generated recs. If any twitter data exists, they could assign a weight that might be proportional to the number of tweets.

The issue with that approach is largely a performance one, judging from what I know of flightcaster. If they have to issue a search to twitter for the flight info, the user has to wait on it.

If they constantly poll, then there are thousands of flights and they have to poll them regularly to make sure they have the latest info on the flight.

IIRC, the current algorithms include historical ontime percentages and weather, but I don't remember reading about them using the realtime web or airline's websites to gather it. Of course I don't really know what I'm talking about, it's all conjecture.

One would also like to ask if it's even true that everyone says this. Are there data to support that claim? :) (Yes, it's just a joke, yet... it would make nice irony.)