Causal graph neural networks for airline profitability prediction

Abstract

This paper employs methods from the field of causal AI to model airline profitability. By modelling causality rather than correlation, complex interrelationships between end- and exogenous factors and airline profitability can be described. Causal models represent a transition from the black-box models of traditional machine learning to a model with better transparency and improved generalisation power, enabling application of the model to new regions and implicit explainability where the model diverges. By implementing a model focusing on causal AI, the aim is not only to increase the accuracy of modelling past data over the entire time span but also to address the possibility of accurately and understandably predicting an airline’s profit based on current data. Through quantification and maximization of the causal effects between the explanatory variables, the understanding of the true mechanism behind airline profit should be increased, and a prediction of future profits based on current figures becomes possible.