Downloads & Free Reading Options - Results

In this thesis, we cluster stop points into stop-point regions using one month’s Automatic Identification System (AIS) data from the Gulf of Mexico and Caribbean Sea to characterize vessel behavior in an area with diverse traffic patterns. Initial cleaning of the dataset is necessary to address multiple issues common to AIS transponders. We consider methods for computing inter-point distances. In particular, we study a promising method for combining geospatial coordinates with other vessel attributes. We use the Ordering Points To Identify the Cluster Structure (OPTICS) clustering algorithm because it can identify outliers, and it constructs clusters of varying shapes and densities. Our best results come from dividing the area of interest into seven zones of equal size, and analyzing the results over each zone. Using classification trees to develop a classification tool, we illustrate an approach for predicting the cluster membership of a new observation. Due to the reduction in computation time and accuracy of results, we recommend that further research utilize the methods from this study as the foundation for an automated threat detection system.

In this thesis, we cluster stop points into stop-point regions using one months Automatic Identification System (AIS) data from the Gulf of Mexico and Caribbean Sea to characterize vessel behavior in an area with diverse traffic patterns. Initial cleaning of the dataset is necessary to address multiple issues common to AIS transponders. We consider methods for computing inter-point distances. In particular, we study a promising method for combining geospatial coordinates with other vessel attributes. We use the Ordering Points To Identify the Cluster Structure (OPTICS) clustering algorithm because it can identify outliers, and it constructs clusters of varying shapes and densities. Our best results come from dividing the area of interest into seven zones of equal size, and analyzing the results over each zone. Using classification trees to develop a classification tool, we illustrate an approach for predicting the cluster membership of a new observation. Due to the reduction in computation time and accuracy of results, were commend that further research utilize the methods from this study as the foundation for an automated threat detection system.