Air Traffic Delay Prediction Based on Machine Learning and Delay Propagation

Meng Li, Purdue University


Flight Delay creates significant problems in the current aviation system. Methods are needed to analyze the manner in which delay propagates in the airport networks. Traditional methods are inadequate to the task. This paper presented a new machine learning based air traffic delay prediction model that combined multi-label random forest classification and approximated delay propagation model. To improve the prediction performance, an optimal feature selection process is introduced and demonstrated to have better performance than directly using all the features of available datasets. Departure delay and late arriving aircraft delay are shown to be the most critical features for delay prediction. To utilize these two features, a delay propagation model is proposed as a link to connect them to build a chained delay prediction model. Given the initial departure delay, the chained model is demonstrated to have the ability to predict the flight delay along the same aircrafts itinerary. By updating the actual departure delay with the iteration number along with the itinerary, the model’s accuracy can be further improved. Our application results demonstrate the value of machine learning and delay propagation for analyzing and predicting the air traffic delay in daily operation.




Sun, Purdue University.

Subject Area


Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server