Machine Learning to predict cetacean behaviour using social and environmental features

Understanding how behaviour is influenced by the environment in which animals move and being able to predict it from social and environmental features is a very ambitious but necessary goal for their conservation. This is especially true for cetaceans, marine mammals with very complex social life and ecology which are nowadays facing various threats to their wellbeing and survival. To that aim, we propose a Machine Learning framework, based on Random Forest and RUSBoost algorithms, to predict cetacean behaviour from a plethora of 27 variables, including the group size and oceanographic features provided by Copernicus Marine Service (CMS). Models have been developed using behavioural data collected in the 2016-2021 period on striped, Risso's and bottlenose common dolphins sighted in the Gulf of Taranto. The performance reached with the ML approach for the classification of feeding behaviour is remarkable, with a prediction accuracy achieved by the dedicated models of about 75%. Thus, the proposed strategy can be successfully implemented in future works to forecast target species behaviour and to investigate further the influence of anthropic variables and other habitat characteristics on it in order to enhance their conservation.