Modelling Flight Delays in the Presence of Class Imbalance

Flight delay is one of the common problems faced by many air passengers. Delays in flights not only bring about inconvenience to passengers, but also cost the airlines. To streamline travel experience, airlines have been leveraging on data analytics to predict flight delays. Although many prediction models have been proposed, they perform poorly especially on data that have imbalanced class distributions. Often, these models pay less attention to the minority 'delay' class, which are usually more relevant and important. In this paper, we address the issue of imbalanced class distributions to improve the overall classification performance in predicting flight delays. We validated our approach on a public airline on-time performance dataset.