@misc{1786,
  keywords = {mywork},
  author = {Chen Zhang and J{\"u}rgen Hackl},
  title = {Sequence-on-Graph Prediction for Transportation Trajectories Using Higher-Order Networks},
  abstract = {Modeling human mobility on networks requires predictive models that are accurate, robust, and interpretable. However, existing approaches fall short: deep learning models, while powerful, often lack these practical qualities, while standard graph learning models can miss sequential patterns. We address this gap with a framework using higher-order networks (HONs) that statistically grounds sequential dependencies to the road network, providing a transparent and topologically consistent state space. In contrast to the parameter-rich architectures of deep learning models, HONs provide a simpler, statistical representation that captures mobility behavior directly on a graph. We evaluate HONs on next-step prediction and robustness to data sparsity against deep sequence model baselines. Our results demonstrate that by modeling sequential dependencies on topological constraints, HONs are able to achieve competitive predictive performance and increased robustness to data sparsity. This work establishes a simpler, networkaware statistical approach as a powerful and transparent alternative to sequence learning models, providing a foundation for understanding human mobility patterns.},
  year = {2025},
}