From Data to Delay Prediction: A study of U.S. Flight Operations in 2024

Abstract

This research explores the phenomenon of flight delays in the U.S. air transport system during 2024, combining a descriptive and explanatory analysis with the development of predictive models. The study first examines the operational context and delay patterns using a dataset of over 7 million commercial flights, identifying key factors such as late aircraft arrival, carrier related issues, and airport congestion. Through a comprehensive review of the literature an industry insights, the project highlights the complex, interconnected nature of air traffic operations and delay propagation. Based on these findings, machine learning classification models are implemented to estimate the probability of arrival delays exceeding 15 minutes. Variables used include temporal, spatial, and operational features, as well as historically aggregated delay data by airline and origin airport. Although the predictive performance of the models is modest, particularly in terms of F1 Score, the results offer valuable insights into which features most influence flight delays. Recommendations include resource optimization, improved airport procedures, and enhanced data-sharing among stakeholders to foster a more collaborative and efficient air travel ecosystem.