Data-driven AIS trajectory reconstruction framework for reliable ship trajectory prediction under sparse and irregular data conditions

This study proposes a data-driven trajectory reconstruction framework for highly corrupted datasets of Automatic Identification System (AIS) data, where high traffic density and complex maritime environments often result in large missing intervals and data anomalies. The framework integrates statistical filtering and ship motion kinematics to identify and remove abnormal data, followed by a hybrid quadratic-cubic interpolation method to effectively reconstruct highly corrupted AIS datasets. Results show that conventional cubic spline interpolation produces physically unrealistic vessel behaviour when applied to highly corrupted data, particularly near trajectory boundaries. The masking experiment shows that the proposed interpolation method achieves improved accuracy with lower errors. Using the proposed framework, the anomaly rate is reduced from 79% to 18% while retaining approximately 80% of the dataset. The reconstructed data are further evaluated using multiple deep learning models for ship trajectory prediction. The proposed framework achieves at least a 15% improvement in root-mean-square error compared to kinematic interpolation under comparable conditions. The quadratic-cubic method consistently achieves lower prediction error, outperforming both standard cubic spline and kinematic interpolation. These findings highlight the necessity and effectiveness of the proposed framework in processing corrupted AIS data for ship trajectory prediction in complex traffic areas.