Combined use of two supervised learning algorithms to model sea turtle behaviours from tri-axial acceleration data

Area of Science:

• Marine Biology
• Animal Behavior
• Biologging Technology

Background:

• Accelerometers are crucial sensors in animal-attached technology for monitoring posture and movement.
• Understanding sea turtle behavior is challenging due to their elusive nature in the wild.

Purpose of the Study:

• To validate the identification of sea turtle behaviors using accelerometer data.
• To assess the accuracy of supervised learning algorithms (Random Forest and CART) in classifying behaviors from accelerometer signals.

Main Methods:

• Tri-axial accelerometers were deployed on juvenile loggerhead, adult hawksbill, and adult green turtles.
• Behavioral data was recorded simultaneously using fixed cameras for validation.
• Supervised learning algorithms (Random Forest and CART) were applied to accelerometer data.

Main Results:

• The Random Forest model achieved 86.96% accuracy for adult turtles and 79.49% for juvenile loggerheads.
• The CART model achieved 81.30% accuracy for adult turtles and 71.63% for juvenile loggerheads.
• Algorithms identified 10-12 distinct behaviors, with refinement of models by analyzing decision rules.

Conclusions:

• Accelerometer data, analyzed with machine learning, can accurately identify sea turtle behaviors.
• This validated approach offers a non-invasive method for studying marine reptile ethology.
• The methodology can be extended to other captive sea turtle species and potentially wild populations.