Cluster-based Visualization of human element interactions in marine accidents.

This study addresses human elements in marine accidents that cause many deaths, environmental pollution, and economic losses. The aim of this study is to identify various human elements that cause marine accidents and to visualize their interaction with each other on a two-dimensional plane. For this purpose, a two-stage study was conducted: the first was a pilot study in which the performance of different dimensionality reduction and clustering algorithms was tested, and the second was the phase in which 2050 textual expressions corresponding to human elements in 871 marine accidents were analyzed with the algorithms featured in the pilot study. The textual expressions were converted into vectors with the BERT model, dimensionality reduction was performed with UMAP, and clustering was performed with the k-Means algorithm. The results showed that the most common type of error and the one that interacts the most with others are navigational errors. It has also been observed that a cause-effect relationship can be established between the clusters such as " fatigue" -" cognitive dysfunction " and "f alse observation "-" violation of navigation rules ". This study contributes to taking more specific safety measures as it directly reveals human elements specific to marine accidents rather than using generic human error classifications. • As a result of dimensionality reduction and clustering analysis, 19 different human elements were detected. Navigation errors have been determined to be the most common type of error. • By using marine accident reports, marine-specific errors were easily detected. • The relationships between the elements are also shown with graphics. It has been determined that there is both a visual and semantic cause-effect relationship between the clusters " Incompetent Engine Room Staff " - " Poor Maintenance and Equipment Failure ", " Fatigue " - " Cognitive Dysfunction " and" Perceptual Error " - " Violation of Navigation Rules ". • The error graph created as a result of the analysis visually highlights the interaction between errors. [ABSTRACT FROM AUTHOR]