Love thy neighbour: automatic animal behavioural classification of acceleration data using the K-nearest neighbour...

Area of Science:

• Movement Ecology
• Bio-telemetry
• Machine Learning in Ecology

Background:

• Biotelemetry, particularly accelerometers, aids in studying elusive species' behavior.
• Manual inspection of high-frequency accelerometer data is impractical for behavior classification.
• Existing machine learning methods for accelerometer data are complex and obscure decision-making.

Purpose of the Study:

• To present an accessible machine learning method for automatic accelerometer data classification.
• To enable researchers with limited computational skills to analyze animal behavior.
• To classify behavioral modes using raw accelerometer data without summary statistics.

Main Methods:

• Utilized the k-nearest neighbour (KNN) machine learning algorithm.
• Applied KNN to raw accelerometer data, avoiding summary statistics.
• Implemented the method using the freeware program R for accessibility.

Main Results:

• Successfully classified 5 behavioral modes across 8 diverse species (quadrupedal, bipedal, volant).
• Achieved accuracy and precision comparable to more complex machine learning methods.
• Provided R script to facilitate the application of the KNN method.

Conclusions:

• The KNN method offers a practical and effective solution for automatic behavioral classification from accelerometer data.
• This approach lowers the barrier to entry for using advanced analytics in animal behavior research.
• KNN can be integrated with positioning methods like GPS for comprehensive movement ecology studies.