Unsupervised Machine Learning as a Tool for Exploratory Analysis of Acoustic Telemetry Data: A Case Study With Northern Pike in Toronto Harbour

Abstract

Spatial ecology aims to further knowledge of an organism's relationship with its environment and guide decision-making related to conservation. The advancement of biotelemetry has facilitated this goal, however, data management, from its acquisition to its utilization, is central to its success. Standard analysis may include separating tagged individuals into predefined groups based on biometrics or capture location, and then comparing relationships among groups, environmental measures, and their seasonal habitat choices. While effective in that it informs on the relationship among variables, this approach is computationally intensive, and the insight provided is limited to behaviour among predefined groups. This study effectively and efficiently leverages machine learning methods - hierarchical clustering and principal component analysis - to explore animal behaviour, thus providing an efficient, alternative method to analyzing acoustic telemetry data. A by-product of this project is software development that can facilitate analysis of acoustic telemetry data.