Resource-efficient data transmission for WiFi-capable bio-loggers based on machine learning

Area of Science:

• Animal Behaviour and Ecology
• Machine Learning Applications
• Wildlife Monitoring Technology

Background:

• Bio-logging devices collect crucial data on animal behavior, particularly for elusive species.
• Modern bio-loggers use WiFi for high-resolution data but consume significant energy.
• Reducing energy expenditure is key to extending the operational life of these devices.

Purpose of the Study:

• To investigate energy-saving methods for WiFi-enabled bio-loggers.
• To implement machine learning for on-board data filtering based on animal behavior.
• To evaluate the effectiveness of decision trees in reducing data transmission costs and energy consumption.

Main Methods:

• Utilized machine learning, specifically small decision trees, to recognize animal behavior from sensor data.
• Trained and evaluated decision trees using a controlled dataset.
• Applied decision tree filtering to a state-of-the-art WiFi bio-logger, the WildFi tag.
• Developed a complete pipeline from data collection to deployable software.

Main Results:

• Decision trees effectively filtered data for WiFi bio-loggers, significantly reducing transmission time.
• Energy savings of 14.68% were achieved in realistic scenarios.
• The approach demonstrated practical gains for animal behavior data collection.
• The use of gyroscopes was found to be highly beneficial for on-board behavior detection.

Conclusions:

• Decision trees are a highly beneficial tool for filtering data in WiFi bio-loggers.
• This machine learning approach offers a promising method for enhancing bio-logger longevity.
• The study supports more efficient and sustainable wildlife monitoring practices.
• Off-the-shelf solutions can be effectively used for practical gains in bio-logging.