Master's Thesis presentation, Oscar Birging and Emil Grimheden

World trade heavily relies on maritime transportation, contributing up to 90% of all shipped goods. However, this sector is still struggling to achieve adequate visibility. Therefore, this thesis seeks to lay a foundation for precise event detection for automatically classifying ships' arrivals to port, decreasing the risk of human errors. A DSC, Data Stream Clustering, model is proposed to work on streamed maritime data in a memory-efficient way. This approach utilises a Landmark window along with BIRCH and DBSCAN during the online phase for clustering and identifying outliers. The identified clusters are stored as convex hulls and are combined with previous clusters in a subsequent offline step. After analysing the results, it was found that the cluster quality was satisfactory when compared to a batch clustering model using the same data. However, the DSC model had difficulty finding sparse clusters, resulting in a TPR of approximately 80% compared to the batch model. On the other hand, the DCS model produced a PPV of about 100%, indicating that the formed clusters were very precise. The results seem promising, but the model has not converged. To reach convergence, the model needs some slight changes in parameters and more training to maximise precision.