Vessel rendezvous, defined as coordinated meetings between ships out at sea, have long been identified as potential indicators of illicit activities such as smuggling or unauthorized transfers. Detecting and predicting these events is difficult, partly because there is a lack of ground truth datasets of known rendezvous, but also because they can involve any type of vessel, with all kinds of movement patterns. This makes it challenging to develop general-purpose models that can identify incoming vessel rendezvous in a natural setting.

To determine whether this is feasible, this study investigates whether models trained only on recent AIS (Automatic Identification System) data can support the prediction of rendezvous events. The first part of the study evaluates the performance of a general short-term trajectory predictor trained across various vessel types and routes. The goal is to understand how well such a model performs in an unconstrained setting and whether its predictions could be useful for further tasks like rendezvous forecasting. The second part focuses directly on predicting rendezvous: a clustering algorithm is used to identify likely rendezvous events, and a deep learning classifier is trained to predict whether two vessels will meet, based on their recent movements and interactions. Together, these two approaches provide insight into the feasibility of real-time rendezvous prediction in diverse maritime environments.