Density Analysis Based Flight Delay Prediction with Genetic Algorithm Hyperparameter Tuning

Abstract

Flight delays are a major problem in the current aviation system. Once there is a flight delay, this causes chain delays at multiple airports, which results in tremendous economic loss. This paper presented a machine learning (ML) approach for the prediction of flight delays by using a decision tree based algorithm called gradient boosting (XGBoost). The genetic algorithm was used as a tuning parameter for optimization to improve the prediction performance. Artificial intelligence (AI) analysis determined the likelihood of the effect of the flight delays by intervals. The departure delays and late arrival delays were the most important parameters for predicting the delays. The result (prediction accuracy) of the density-based method model was compared to the long short-term memory (LSTM) method. The experimental result showed that the top three causes of delays were the late arrival of aircraft by more than 45 minutes, current delay status of the airport, and the amount of planned departure flights.