An Improved SVM Model for Flight Delay Prediction

Flight delays are aggravated by the increasing number of flights. Although it cannot be avoided completely, early warning through accurate prediction is an effective way to mitigate the impact of delays. There has already been numerous studies on delay prediction, but still exists room for improvement. The selected feature factors can be further optimized. And the daily periodicity of flight plans should not be ignored. Besides, changes in hourly flight delays in historical data also ought to be concerned. In this paper, an improved Support Vector Machine (SVM) model is established to predict the flight departure delay. In the feature selection process, a more comprehensive consideration of the factors affecting flight delays is introduced from three major aspects: the airports, airlines and aircraft. For reducing model complexity, Principal Component Analysis (PCA) is adopted to achieve dimensionality reduction. The typical flight plan data of each flight is filtered into the training set so as to avoid tag jumping. Furthermore, thanks to the periodicity of flight plans, the number of departure flights and departure delay rate in historical data are provided to SVM as prior knowledge. Finally, the prediction model is validated using historical data of the Beijing Capital International Airport.