Development and application of a machine learning algorithm for classification of elasmobranch behaviour from...

Area of Science:

• Animal behavior analysis
• Machine learning applications in ecology
• Marine biology

Background:

• Understanding animal activity budgets is crucial for ecological insights.
• Accelerometers provide high-resolution data for studying animal behavior.
• Automated behavioral classification is needed due to large data volumes.

Purpose of the Study:

• To assess machine learning (ML) classifier performance for discerning five behaviors in juvenile lemon sharks (Negaprion brevirostris).
• To identify the most effective ML model for automated behavioral classification using accelerometer data.
• To investigate the biological relevance of ML-classified behaviors, specifically headshaking as a proxy for prey capture, in wild sharks.

Main Methods:

• Collected accelerometer data from juvenile lemon sharks in a semi-captive environment.
• Used observed behaviors (chafing, burst swimming, headshaking, resting, swimming) for ground-truthing ML models.
• Trained and tested logistic regression, artificial neural network, random forest, gradient boosting, and voting ensemble (VE) models.

Main Results:

• The voting ensemble (VE) model achieved the highest classification performance (F-measure 0.88), outperforming the best base learner, gradient boosting (0.86).
• Application of the VE model to wild shark data revealed significant relationships between headshaking (prey capture proxy) and time of day, tidal phase, and season.
• Prey capture events were most frequent in the early evening and least frequent during the dry season and high tides.

Conclusions:

• Machine learning, particularly the voting ensemble model, provides an effective method for automated behavioral classification in sharks using accelerometer data.
• The study validates the use of ML-derived behavioral data for ecological research, supporting previous hypotheses on shark predation patterns.
• Findings highlight the influence of environmental factors (time of day, tide, season) on juvenile lemon shark prey capture.