Frequently Asked Questions

How do you come up with the predictions?

We use advanced predictive models – essentially sets of mathematical equations – which attempt to represent the patterns seen in historical flight delays. Once we know that mathematical relationship, we use those models to look forward and project those patterns into the future. Translation: Lots of data, lots of complex math and lots of processing power.

How far out do your predictions go?

We look out four to five months.

How often are the models updated?

We refresh our historical data monthly. However, we may fine tune our models where we see performance gaps, so updates can come at any time.

Why do I get a prediction for a time where an airline does not have, and has never had actual flights?

We designed our models to extract patterns about carriers and airports, rather than specific flights. In this way we can predict a broad range of possible flights without being subject to the constantly evolving carrier schedules. We hope to cover as many possible flight segments and times as we can. 
 

On the day I fly there may be weather which can cause substantial delays, yet you don't take the weather forecast into account.

These are planning tools which show you that, even before the effect of weather and mechanical problems, you may be at a disadvantage.  In short, these are the predictions of delays before those factors even kick in.

How is this better than the historical flight performance statistics I get at my travel site?

There are several differences.

1. Historical statistics are backward-looking. Our predictions are based on conditions we expect to see in the future.
2. Historical statistics are simple averages over some fixed time (e.g. the last thirty days). Our predictions are specific to the day and time that you are interested in, and take into account factors like the day of the week, holidays, developing air traffic trends, and other relevant factors.
3. Historical statistics are flight-specific. Our models can predict many newly launched day/time/airline combinations. So, when an airline changes the time of a flight or adds a new flight (schedules are frequently updated), we you are covered.  It will take the historical, backward-looking approach at least a month before it can start providing you its averages.
4. We have yet to find a site that tells you about arrivals, not just departures.
5. Finally, and most importantly, our predictions will more often provide you with a more accurate estimate of the delay potential of your considered flights.

My flight was on-time but you said it would be so many minutes late.  What gives?

Hopefully you will be on-time most of the time! The prediction of your delay length in minutes assumes your flight is delayed in the first place. See the predictions called “chance you your flight departing/arriving on-time.”  Many flights do make it on-time, but if they don’t, the minutes delayed section tells you what to expect in terms of the length of the delay.

Bear in mind, too, that these are estimates. Sometimes your flight will be on time even if we say it has only a 20% chance, and sometimes the delay will be 60 minutes even if we predicted 40. And always be aware that other factors such as weather may further delay your flight beyond our so-called “base expected delay” prediction.

Why is there a difference between the departure and arrival predictions for a flight segment/time I am considering?

There will almost always be some difference. Here are several reasons for the gap:

• Flights can officially depart the gate a certain number of minutes late but experience further delays on the tarmac or in the air.
• Flights can make up time in the air (in fact, slack is built into the schedules).
• The predictions are “fuzzy” in the sense that the precision is not exact. Small gaps between departure and arrival predictions are expected.

If you see what you believe to be a large discrepancy, email us.

What airports and airlines do you cover?

We currently make departure and arrival predictions for the top 60 airports.  We track the following airlines:

Can you add my airport/airline to the list?

We may expand beyond the current set after our beta period, depending on how solid we can get our predictions at smaller airports where fewer data points are available.

How good are the models?

Good enough to aid decision making. That is the purpose. Frankly, we believe our model provide the most accurate forecast of flight delays available anywhere.

That said, remember that models are representations of reality they are never exactly accurate. In our reviews, we are within 15 minutes 80 - 90% of the time. (If carriers were within 15 minutes of their published schedule 80-90% of the time, there would be no demand for Delaycast, or Congressional hearings.) Of course that varies by airport. We are constantly refining, so as time goes by we expect our forecasts to improve.

Can I purchase large chunks of your prediction data?

Yes. Please email us your anticipated usage volume and application as well as your interest in widgets, APIs or hosted OLAP data.

Can you predict how long it will take me to get through security on the day/time of my flight?

We feel your pain. We are working on it.

Who are you guys?

We are analytical professionals who spend a lot of time waiting in airports and on tarmacs.

We originally developed mental models to help us cope with travel woes and to set realistic expectations about our own travel. One of us actually kept a little travel log of data to run simple statistical models predicting the expected arrival times of his flights. After some conversations and the statisticians' equivalent of a few jam sessions, we came up with some amazingly solid models. From that, Delaycast was born and taken to a large scale.

Mostly, we just like solving some of life's problems with big data sets, elegant equations, and brawny computers. We formed Delaycast, Inc to make this application available to as many people as possible.