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

Bio-logging is a popular method for data collection in animal research, especially for hard-to-observe animals. Newer bio-logger generations utilise WiFi technology, enabling researchers to collect high-resolution data at the cost of higher energy expenditure of the devices. In this study, we elaborate on how state-of-the-art loggers can benefit from even the simplest methods to reduce transmission costs. We employ machine learning techniques, specifically small decision trees, to enable a bio-logger to recognise a chosen behaviour based on its sensor readings. Based on the recognised behaviour, the logger filters which data to transmit, reducing transmission time and, thus, the logger's overall energy consumption. Using a controlled dataset, we exemplify the training and evaluation of such decision trees. Using those, we evaluate the reduction of energy consumption based on a state-of-the-art bio-logger, the WildFi tag. We demonstrate that for WiFi-enabled bio-loggers, decision trees are highly beneficial when used as a data filter. We illustrate that filtering with decision trees yields energy savings of 14.68% in realistic scenarios for transmitting data. We provide a full pipeline from data collection to deployable software to holistically elaborate on how to use off-the-shelf solutions to achieve practical gains for animal behaviour. Our results suggest that decision trees can be an effective tool for enabling bio-loggers to detect specific behaviours. Lastly, we emphasise that our approach highly benefits from the use of gyroscopes, a sensor type that mostly sees use for off-board instead of on-board labour. We contribute an investigation of energy consumption reduction of WiFi-enabled bio-loggers through the utilisation of controlled data transmission using machine learning. We offer a promising pathway for enhancing the longevity of such a state-of-the-art bio-logger, maintaining WiFi benefits. Ultimately, we support more efficient and, thus, more sustainable wildlife monitoring practices on the example of the WildFi tag.