Detection of suspicious behaviour from ship transponder data

Since the 1990s, data of vessels has been collected and sent by a so-called Automatic Identification System (AIS). The data contains information about the vessels such as the speed, the location and the vessel type (for example cargo, tanker or fishing). In this research, we aim to detect suspicious ship behaviour from this AIS data. To do so, data mining techniques are applied to get more insight in this relatively unknown domain. Thereafter, machine learning is applied to predict the vessel type based on 11 features. This prediction is compared with the real vessel type to identify vessels with untypical behaviour: suspicious vessels. The derived prediction model has a F1 score of 0.70 using all features and a F1 score of 0.58 using only dynamic (radar) features. The model can be applied in the real world to perform vessel controls based on facts instead of performing them randomly, which can lead to a more efficient use of resources by the water police.