Vessel Traffic Flow Prediction on Sparse Data via Spatio-Temporal Graph Neural Networks with a Learnable Tweedie Head
Researchers have developed a new plug-and-play output module, the learnable Tweedie head, designed to enhance spatio-temporal graph neural networks (ST-GNNs) for predicting vessel traffic flow. This module specifically addresses the challenge of sparse and intermittent maritime data, which often causes conventional ST-GNNs to produce overly conservative predictions. By optimizing the Tweedie unit deviance and learning node-level variance, the new head improves forecasting accuracy, particularly for non-zero events, as demonstrated in experiments using real-world AIS data from the Ports of Los Angeles and Long Beach. AI
IMPACT Enhances forecasting accuracy for sparse maritime data, potentially improving smart port operations and navigational safety.