Predicting Flight Delays by Using a Meta-classifier

Airlines around the world are facing delays which incur extra costs to airlines, passengers, and society and difficulties for airport operations. Thus predicting the occurrence and magnitude of these delays is a major objective of airlines. This paper identifies the most important factors that generate delays in airline operation. It also provides a new hybrid method to predict flight delays. This technique, called meta-classifier, combines two conventional machine learning approaches: the decision tree approach and the cluster classification approach. Proposed technique was implemented on an actual dataset of one of the major Iranian airlines with more than 35 airplanes that fly between 52 airports. Results show age of fleet, aircraft type, and departure time have the most influences on flight delays. Proposed hybrid approach has more accurate performance compare to regular data mining techniques. For this method, the accuracy of delay occurrence prediction is 76.4 percent and the accuracy of estimation of delay magnitude is 76 percent. It is hoped that this technique will enable airline agencies to predict delays accurately and improve flight planning process to prevent delay propagation.