APPLICATION OF INTERPRETABLE MACHINE LEARNING IN FLIGHT DELAY DETECTION

Abstract

Precise flight delay prediction is vital for the airline industries and passengers. This thesis focuses on applying several machine learning and auto-ML techniques to predict flight delays. A flight delay is said to occur when an airline lands or takes off later than its scheduled arrival or departure time, respectively. Conventionally, if a flight's departure time or arrival time is greater than 15 minutes than its scheduled departure and arrival times respectively, then it is considered that there is a departure or arrival delay with respect to the corresponding airports. Notable reasons for commercially scheduled flights to be delayed are adverse weather conditions, air traffic congestion, a late reaching aircraft to be used for the flight from a previous flight, maintenance, and security issues. In this research study, a python-based model will be developed for a specific Airline and an Airport from already established models that are available in literature and were implemented in flight delay predictions. Once that is completed, the same model will be used for a different Airline at the same Airport. Later, the model will be implemented for several other Airports to check the adaptability of the models. In this process, there will be an attempt to enhance the existing models by carefully selecting the dataset and features. In the final stage, the results will be compared with the Microsoft Azure Machine Learning Studio, the best model will be deployed using Auto-ML and the existing interpretable machine learning package, LIME will be used to explore local prediction capability of the models. This study has been conducted with the hopes that alongside other increasing numbers of studies in this subject matter, it will contribute to improving on-time performances of flights to benefit airline customers, airline personnel, and airport authorities.