Abstract: Currently, flight delay is a growing serious problem worldwide, so the task for forecasting the warning level of flight delays is more and more pressing. Based on the viewpoint that flight delays result from the conflict between the air side demand and airport capacity, this article first introduces the ranking support vector machine (Ranking SVM) for forecasting the warning level of flight delays; secondly, an incremental algorithm of the Ranking SVM is proposed to adapt to the specific characteristic that data of airport flights have to be updated ceaselessly. Experiments on toy data sets show that the incremental algorithm is more satisfying to the need of online forecasting than the present method. Moreover, predicting accuracy reaches 80% on the flight dataset collected from an international airport in China about two weeks ahead of the flights.

Key words: warning level of flight delays, airports, support vector machines, ranking learning, incremental learning