Integrating multiple data sources for improved flight delay prediction using explainable machine learning

Flight delays negatively impact costs, customer satisfaction, and revenue in the aviation industry. As a result, it is critical to identify the factors that cause flight delays for each airport, as they can vary depending on various attributes associated with their operations.

This study proposes an explainable artificial intelligence (xAI) methodology for identifying the features that affect airport delays by integrating data from multiple sources and implementing explainable artificial intelligence. The methodology incorporates operational data, airport information, geographic data, and weather data combined and used to train a series of machine learning models. Furthermore, the SHAP and Sobol techniques are used to thoroughly analyze the features that influence flight delays for the specific case of the airport in Santiago, Chile.

The results show that a linear discriminant analysis model is best suited for predicting flight delays in this specific case study, and the features that have the most significant impact on delays are the international flight status, average temperature at the destination airport, wind speed, and average temperature at Santiago airport.

The proposed methodology could be applied by airlines that can collect data from multiple sources and conduct similar investigations, leading to the development of a decision support system to make better-informed decisions and reduce the impact of flight delays.