Deep learning for risky encounter prediction: Leveraging LSTM networks and AIS data

The maritime industry relies heavily on large-capacity vessels, raising concerns regarding the reliability, safety, and stability of maritime transportation. Predicting risky encounters is crucial for enhancing safety and efficiency. In this paper, we introduce a practical framework for predicting risky close encounters and gray-zone interactions using historical Automatic Identification System (AIS) data and Long Short-Term Memory (LSTM) networks. Our methodology develops an extensive encounter model to identify ship-to-ship interactions and classify them as risky, gray-zone, or non-risky encounters based on ship domain violations. The LSTM networks predict the likelihood of hazardous encounters and gray-zone interactions based on sequential AIS messages and variables such as ship dimensions, speed and course. Results indicate that our framework shows promise in predicting risky maritime encounters and gray-zone interactions, contributing to improved navigational safety and risk mitigation strategies. The Strait of Istanbul serves as a case study, demonstrating the real-world applicability, scalability, cost-effectiveness, and interpretability of our approach for maritime authorities, emphasizing the potential of deep learning and LSTM models in maritime collision risk prediction.