Identification of Cassin׳s Vireo (Vireo cassinii) individuals from their acoustic sequences using an ensemble of learners

The ability to identify individual birds can provide us with information on behavioral interactions between individuals and ecological interactions with the environment. Identifying individual animals has traditionally been a time – and cost – intensive exercise in the field. Most past efforts dedicated towards individual identification from its vocalizations have centered on the analysis of acoustic features. Here we present an alternative approach to this task, using an ensemble of learners to identify individual Cassin׳s Vireo from the structural properties of their vocalizations, using symbolic representations of its syntactic elements instead of the acoustic characteristics of the signal. We also test the ability of this ensemble of learners to identify individuals within a year and across years. We propose a new learner combination that confers the ensemble with the ability to handle outliers – unknown individuals not seen during training. After being trained with 9 individuals from one year (2014), the ensemble achieved 96% accuracy identifying samples from 13 individual birds from the same year and 95% with sequences from 8 individual birds from a previous year. Predicting individuals in one year using recordings from other years indicates effective generalization capabilities of the ensemble.