Chaos amidst the Flock: Clustering Oil Barrel Trajectories

Abstract

This study explores the effectiveness of various clustering algorithms in reducing data variability in maritime oil barrel pollution analysis. The research investigates four clustering methods - k-Means, Agglomerative, HDBSCAN, and OPTICS - applied to simulated trajectories of oil barrels. The focus is on assessing the reduction in data variability using standard deviation and Mean Absolute Deviation (MAD) as reduction rates. The findings demonstrate that density-based clustering methods, particularly OPTICS, significantly reduce data variability by categorizing noise effectively. However, this approach may not be suitable for applications requiring the inclusion of all trajectories, such as identifying offending ships. In these scenarios, distance-based methods perform better but offer minimal data reduction. These results underscore the importance of selecting appropriate clustering methods based on specific requirements. The broader perspective suggests that while clustering can enhance data analysis efficiency, careful consideration of the trade-offs between data reduction and information retention is essential for reliable maritime pollution tracking.