Home Blog Archives RSS
¶ learning airline ticket prices
Another idea I think is cool, but will probably never get around to doing. This one is somewhat more speculative. It's also not my idea - this one goes out to Dan Weld out at the University of Washington, who already did it to a first approximation, I think... which makes me wonder why I'm writing about it in the first place... oh well.

Unlike bus and train fares, air travel ticket prices fluctuate dramatically in a seemingly random fashion. A seat on the same flight from Boston to San Francisco can change price by hundreds of dollars in a matter of days. People are often left wondering whether waiting a day or two would get them a better price, or if they should buy immediately.

These prices are not completely random. A friend in the industry tells me that there are hordes of people whose jobs are to sit at a terminal and change ticket prices. I don't know the exact methodology, and there are probably rules and spreadsheets that dictate how the prices are to change, but at the end of it all, there's a human in the loop setting the price.

What if you could train a system to learn when is the best time to buy a ticket? You tell the system the time and endpoints of your planned flight, and then it tells you when you should buy your ticket. Machine learning algorithms are strong enough now to detect complex patterns in ticket prices and finding a local minimum is a simple matter. Dan told me he did this a few years ago and got good results - don't remember the details, though. You could probably formulate this as a Bayesian inference problem and train it on lots of old data.

So how much would you pay to (probably) save a couple hundred dollars?

No comments, be the first!

Comments disabled until the spammers go away. I hope you comment spammers all die horrible deaths and are forced to delete endless streams of comment spam in your days in purgatory.
• Powered by bBlog