Google Flights Will Now Predict Airline Delays -- Before the Airlines Do (techcrunch.com)

Google is rolling out a few new features to its Google Flights search engine to help travelers tackle some of the more frustrating aspects of air travel -- delays and the complexities of the cheaper, Basic Economy fares. From a report: With the regard to delays, Google Flights won't just be pulling in information from the airlines directly, however -- it will take advantage of its understanding of historical data and its machine learning algorithms to predict delays that haven't yet been flagged by airlines themselves. Explains Google, the combination of data and A.I. technologies means it can predict some delays in advance of any sort of official confirmation. Google says that it won't actually flag these in the app until it's at least 80 percent confident in the prediction, though.

Comments

Probably not that hard to do better

[Nkwe]

I fly on United regularly and one of my frustrations is that the airline is slow to update the status of flights that they should know are going to be late. Using their own smartphone app (or website) you can see the status of a given flight and can also easily navigate to the previous flight leg for where your plane is coming from. I have observed that it is very common for my flight to be listed as "on time" when the the app shows that the plane I am scheduled to be on is going to be late in arriving to the airport (making it physically impossible for my flight to be on time.) I get that the schedules have quite a bit of padding and that a short delay in an inbound flight doesn't necessarily mean that the outbound flight will end up be late in its ultimate destination, but when the airline knows that the inbound flight is going to be an hour late, it's shouldn't take half an hour for it to update the status of all the cascading delays. Perhaps Google can be more aggressive at prediction or maybe at least create some prediction competition so that United gets more timely with its updates.

Planes are pretty full these days and knowing 15 minutes earlier that you are going to miss a connection can easily make the difference as to if you can make other arrangements or not.

Machine Learning can do a lot better

[Roger W Moore]

Some of the airline reservation services display an "ontime" percentage next to the flight - EG This flight is ontime 75% of the time. You don't need machine learning for something that simple statistical analysis will do.

You do if you want to provide better information than this. For example, I would be fine taking a flight which is 75% on time and the remaining 25% of the time is only 20 minutes late. I would be less happy if 25% of the time the flight was 2 hours late. Also this statistic ignores patterns in the delays. For example, suppose the flight is on time 90% of the time from Mon-Thu but delayed by 2 hours 85% of the time on Friday? This would be completely consistent with the data provided if you average over the week but clearly, you would never want to take this flight on Friday!

While you might be able to do a simple analysis on a weekly pattern airline schedules are more complex and if the delay is caused by a complex interaction of the crew, plane and airport schedules it will be a lot harder to pick up manually whereas a machine learning algorithm should easily be able to cope with this level of complexity and spot that on the third tuesday and second wednesday of every month your flight is delayed. In fact, if the airlines do not do this already, this might be a useful tool for them to spot and identify problems so they can fix them.

Re:Seems easy....

[dgatwood]

I don't understand why DFW would be a problem ...

During certain times of year, winds shift in such a way that they go from using the five parallel north-south runways (four primary, plus one shorter runway for small planes) to using the two 45-degree runways.

When this happens, they have to stop all departures and all arrivals that aren't already lined up, and once the inbound traffic is cleared, then and only then can they resume takeoffs and landings on the new runways. IIRC, the changeover itself creates about a twenty or thirty minute hiccup every time they switch runways.

But to make matters worse, at that point, there are less than half as many active runways, which reduces the maximum traffic volume proportionately. So if it stays in that configuration during high-volume periods, delays can pile up, and in some cases, they may end up diverting some arrivals to Love Field.