Abstract: Flight delay prediction is significant for busy airports. Aiming at the difficulty of predicting the number and duration of delays in busy airport flights, SVM(support vector machine) regression method is used to establish the flight arrival/departure delay prediction model. First of all, according to the flight operating data, data mining and backward stepwise selection algorithm are used to determine the most relevant factors of number and duration of delay per hour respectively. Secondly, grid-search and cross-check methods are used to select the optimal model parameters. Finally, historical data of LAX(Los Angeles International Airport)and PVG(Pudong International Airport)are used to train the model, and multivariate linear regression model and SVM regression model are applied to test the current model. Comparison results show that the SVM regression model can achieve better prediction effect.

Key words: flight delay, SVMregression, backward stepwise selection