FLIGHT DELAY PREDICTION BASED ON AVIATION BIG DATA AND MACHINE LEARNING

Abstract

Accurately predicting flight delays is crucial for optimizing airline operations. While prior research has primarily focused on single routes or airports, this study takes a broader approach by considering various factors influencing delays. By integrating automatic dependent surveillance-broadcast (ADS-B) messages with weather conditions, flight schedules, and airport data, a comprehensive dataset is constructed. The prediction tasks include classification and regression, with experiments revealing that long short-term memory (LSTM) models struggle with overfitting due to dataset limitations. However, a novel random forest-based approach demonstrates superior performance, achieving higher prediction accuracy (90.2% for binary classification) while effectively addressing overfitting issues compared to previous methods.