Follow Datanami:
April 8, 2013

Drawing an Optimized Flight Plan with GE

Ian Armas Foster

According to GE, commercial airlines spend about $22 billion every year managing flight plan efficiency. To help curb those expensive costs, GE put out what they call the “Flight Quest Challenge,” encouraging data scientists and analysts to crunch flight plan numbers and come up with better algorithms.

In the video below produced by GE, 2nd prize winners Jonti Peters and Pawel Jankiewicz discussed the goal and the challenges of the competition.

“The end goal of the competition is to come up with a model that can be used in-flight so planes can know more accurately know when they will land at an airport,” Peters said. The model derived by Peters and Jankiewicz, who together formed “Team As High As Honor,” combined a generalized linear model with a random forest model to refine an existing linear model. These refinements allowed them to fill in variables missing in certain data sets.

“The flight competition involved numerous tables with data regarding the flight conditions, weather, estimations of arrival times, and we had to combine this all together,” said Jankiewicz on some of the variables they needed to consider.

GE ended up handing out $250,000 total to the top five winners of the Phase One competition. The algorithms ended up increasing predicted flight time arrival by 40 percent. The winning algorithms will, according to GE, be used in the second phase of the Flight Quest competition, the goal of which will be to help build an on-flight management application to increase efficiency.

First prize went to a team out of Singapore who also used random forest models, along with gradient boosting, to help refine the data. Their final model pulled anywhere from 58 to 84 features out of an optimized total set list of 258. It was this algorithm that produced a 40 to 45 percent increase in determining runway arrival times.

 “The recently developed algorithm as part of the Flight Quest challenge gives us another tool that we will continually apply to our processes of improvement,” said Frank Martin, Seattle Director of Station Operations, Alaska Airlines. Indeed, that $22 billion spent on managing efficient flight plans across the country represents a sizable portion, when spread out among airlines like Alaska. It is also where data scientists in big data have one of their larger opportunities to affect the airline industry, as it all revolves around data optimization.

The second phase will reportedly launch on June 30th of this year.

Related Articles

Python Wraps Around Big, Fast Data

Data Athletes and Performance Enhancing Algorithms

Rometty: The Third Era and How to Win the Future