Remotely monitoring maritime vessels is crucial for policing and for ensuring their safety along Australia’s extensive coastline. Detecting unusual behaviour is vital for evaluating naval activity, responding to illegal activities, and detecting ships in distress. AIS (Automatic Identification System) data provides a rich source for research in this domain; however, it only provides a limited set of features and the data is noisy and irregular. In this study, we augment sampled trajectory segments of base AIS data with a suite of statistical features and label them with behaviours in order to perform supervised learning. We evaluate these features in classifying maritime vessel behaviour using three different machine learning classifiers - Random Forests, K-Nearest Neighbour and a Multilayer Perceptron. The latter performed best using a set of the twenty most important statistical features, as determined by Gini Impurity. It returned a weighted-average F1 Score of 83.6%. These results demonstrate that a classifier built from a manually labelled AIS dataset can discriminate vessel behaviour using only simple statistical features that are easy to generate and can be applied to a range of classifiers. This finding lays the groundwork for future research aimed at identifying anomalous behaviour in maritime traffic.