Analyse Near Collision Situations of Ships Using Automatic Identification System Dataset

摘要

Ship-to-ship collision is an area of interest for many stakeholder groups such as ship owners, regulatory authorities and insurers etc. Many researchers have found different methodologies for analysing ship-to-ship collisions. However, analysing ship-to-ship collisions is limited by the number of data points available. As a solution, this study utilises the high data availability in near collision situations since near collisions are more common than the actual ship-to-ship collisions. This paper studies the near ship-to-ship collision situations using Automatic Identification System (AIS) data set. The first part of the study defines a near collision criterion by two existing mathematical and statistical models. Defining this ship domain considers factors such as Speed over Ground (SOG), Course over Ground (COG), width of the ship and the length of the ship etc. In the second part, the study defines the selected ship domain and the near collision criterion with the real-time AIS dataset extracted from the USA coastal area. Application of the ship domain and near collision criterion requires handling a large volume of AIS data. In this study, AIS data of one month on one coastal zone consists of approximately 25 million of data records. And in the final part of the study interprets the descriptive statistics of the identified near collision situations and evaluates four data mining algorithms (C5.0, Support Vector Machines, Naïve Bayes and CART) for predicting the near collisions with the defined criterion.