PROJECT TITLE :

A Deep Learning Approach for Flight Delay Prediction through Time-Evolving Graphs

ABSTRACT:

Because of the significant role it plays in the effective operation of airlines and airports, the ability to predict flight delays has recently been enjoying a surge in popularity. The vast majority of the earlier prediction works take into account the case of a single airport, which ignores the time-varying spatial interactions that are concealed in airport networks. In this paper, we investigate the problem of flight delay prediction from the viewpoint of a network (i.e., multi-airport scenario). In this paper, a method for predicting flight delays is developed that is based on a graph convolutional neural network (GCN). The goal of this method is to model the time-evolving and periodic graph-structured information that exists within the airport network. To be more specific, given that GCN is unable to accept as inputs both delay time series and time-evolving graph structures, a temporal convolutional block that is based on the Markov property is used in order to mine the time-varying patterns of flight delays through a sequence of graph snapshots. In addition, because unknown occasional air routes during an emergency may result in incomplete graph-structured inputs for GCN, an adaptive graph convolutional block has been incorporated into the proposed method in order to unearth spatial interactions that have been hidden in airport networks. Extensive testing has shown that the proposed method outperforms the benchmark methods with a satisfactory improvement in accuracy at the cost of acceptable execution time. This improvement comes at the expense of the execution time. The results that were obtained indicate that a deep learning approach that is based on graph-structured inputs has significant potential in solving the problem of predicting flight delays.