Long-term ship trajectory prediction using a transformer with inverted attention and feature augmentation

Maritime transportation is essential for global trade, with the increasing ship traffic necessitating accurate trajectory prediction for enhanced safety and efficiency. In this study, a transformer-based architecture is proposed for long-term ship trajectory prediction. Feature augmentation is performed by deriving kinematic and directional variables from raw AIS data, and trajectory clustering is applied using dynamic time warping. An inverted attention mechanism is employed, to compute the attention across variables rather than temporal positions, thereby enhancing scalability in high-dimensional settings and enabling explicit modeling of variable dependencies. The encoded representations are mapped to the prediction horizon through a multilayer perceptron decoder. Comprehensive experiments on AIS trajectory datasets demonstrated that the proposed framework attains higher accuracy in both short- and long-term prediction tasks. The results indicate that the integration of feature augmentation and inverted attention enhances predictive accuracy, robustness, and generalization for maritime trajectory prediction.