Classification of behaviour with low-frequency accelerometers in female wild boar

Area of Science:

• Animal behavior analysis
• Wildlife ecology
• Machine learning applications

Background:

• Automated animal behavior sampling offers advantages for behavioral, ecological, and wildlife management studies.
• Low sampling rate accelerometers (1 Hz) are available as ear tags, but their utility for behavior prediction is questioned.

Purpose of the Study:

• To assess the usefulness of low data collection rate accelerometers for predicting animal behaviors.
• To classify the behavior of female wild boar using acceleration data from ear-tag sensors.

Main Methods:

• Wild boar behavior was classified using acceleration data from 1 Hz ear-tag sensors.
• A random forest machine learning model (h2o) was employed for behavior prediction.
• Static features of unfiltered and filtered acceleration data were analyzed.

Main Results:

• Accurate prediction of several wild boar behaviors (e.g., foraging, resting, lactating) was achieved with low-frequency data.
• Prediction accuracy varied by behavior, ranging from 50% for walking to 97% for lateral resting.
• Static features of acceleration data were more important than waveform for prediction; low frequency was sufficient.

Conclusions:

• Ear-tag accelerometers with low sampling rates are effective for automated, non-invasive monitoring of key wild boar behaviors.
• Identified behaviors like foraging and lactation are valuable for wildlife managers, particularly for pest species management.
• The technology supports long-term data collection, enabling the study of seasonal and inter-annual behavioral trends.