Have you had to pick someone up from the airport only to end-up doing several laps of the airport because the flight was late? This sort of setback due to arrival delays is one of those modern day inconveniences that FlightCaster and their FlightCaster API hope to eliminate.
FlightCaster aims to provide accurate predictions about flight delays, often hours before any official notification from the airlines. As the FlightCaster about page says:
We are not regurgitating information from the airline or FAA. You can get that from many many other sources. Instead, we're creating the next generation of flight information by using applied mathematics to provide real predictive power.
While we cannot predict every scenario, such as a freak mechanical failure, our algorithms can find 95%+ of delays.
FlightCaster predictions are currently limited to domestic flights within the US, although Canadian flights and international flights to/from the US are planned for later releases. Anyone wanting to use the service can go to the FlightCaster website and enter their airline and flight number to get a delay prediction.
Data Wrangling have posted an interesting interview with Brad Cross (the FlightCaster Math Guy) where he gives some technical insight into the underlying systems that were used in FlightCaster like Rails, Hadoop, Amazon EC2 and Clojure.
For developers, FlightCaster provides a free REST API, which returns XML results (JSON does not appear to be supported). It is still under development, but has already been successfully used in iPhone and BalckBerry mobile applications. Early adopters will have to watch out for the incomplete documentation though.