Flight Delay Prediction Using a Hybrid Deep Learning Method

Abstract

The operational effectiveness of airports and airlines greatly relies on punctuality. Many conventional machine learning and deep learning algorithms are applied in the analysis of air traffic data. However, the hybrid deep learning (HDL) model demonstrates great success with superior results in many complex problems, e.g. image classification and behaviour detection based on video data. Interestingly, no previous attempts have been made to apply the concept of HDL in analysing structured air traffic data before. Hence, this research investigates the effectiveness of the HDL in the departure delays severity prediction (i.e. on-time, delay and extremely delay) for 10 major airports in the U.S. that experience high ground and air congestion. The proposed HDL model is a combination of a feed-forward artificial neural network model with three hidden layers and a conventional gradient boosted tree model (XGBoost). Utilising the passenger flight on-time performance data from the U.S. Department of Transportation, the proposed HDL model achieves a sharp rise of 22.95% in accuracy when compared to a pure neural network model. However, with current data used in this research, a pure machine learning model achieves the best prediction accuracy.