A novel untapped flight segment flow prediction framework based on graph deep learning and heuristic algorithm for sustainable transport development

Abstract

This study explores a novel machine learning framework for predicting the flow of untapped flight segments, focusing on the unique challenges posed by the absence of historical flow data in airline networks. Utilizing a real-world datasets from a major airline, we evaluate the performance of a graph deep learning-based approach that combines Multi-Graph Attention Networks (MGAT) and Long Short-Term Memory (LSTM) networks, as well as Nondominated Sorting Genetic Algorithm II. The results demonstrate that the proposed framework significantly outperforms traditional models in accurately predicting passenger flow for new flight segments, particularly when compared to statistical benchmarks like time-series models that rely on historical flow data. Moreover, we find that optimizing the affinity coefficients within MGAT using the NSGA- II not only enhances predictive accuracy but also improves the interpretability of the model. Finally, we provide an in-depth analysis of the key factor that influence the predicted outcomes, highlighting the critical role of market competition in untapped segment operations.