An Ensemble Decision Trees Model to Predict Traffic Pattern for Maritime Traffic Management

Abstract

This study presents a traffic pattern prediction model using ensembles of decision trees, leveraging AIS data to classify maritime traffic patterns. The model integrates static information, such as origin and destination, with dynamic data, including ship speed, course and spatial position, to define and extract relevant traffic features. By combining traditional algorithms with a decision tree ensemble model, a stacked predictive framework is constructed and trained on these extracted traffic characteristics. The model is applied and validated using data from the Fujiangsha waters of the Jiangsu section of the Yangtze River. Comparative analysis reveals that this model consistently outperforms traditional algorithms and ensemble models, maintaining stable accuracy above 98% across diverse scenarios. Testing on unseen ship data further confirms the model's predictive reliability, aligning well with actual navigation patterns. The findings suggest that this model has strong potential to (1) forecast navigation routes for improved traffic management, (2) infer ship behaviour based on predicted traffic patterns and (3) support future applications in intelligent ship navigation.