Accelerometers allow the remote study of animal activities and thus offer a huge advantage over traditional studies based on direct human observations. Beyond simple activity, accelerometry can reveal complex behaviours, such as eating and preening. This is done using supervised machine learning methods, whereby a model is trained by annotating accelerometry data with behaviours based on direct observations of tracked birds.
The aim of the project is to build a machine learning model and apply it to study the behaviour of sick birds and to test the hypothesis that repeated exposure to low-dose infection induces infection tolerance. Tolerance is manifest as an attenuated response to infection, and birds that use bird feeders could acquire tolerance if they are exposed to repeated low doses of infections at feeders. Tolerant birds produce a weaker fever and reduced sickness behaviours. This could enable them to continue to be active and subsequently transmit infections to other birds.
Great tits, housed in aviaries at the department’s field station, will be given repeated immune challenges, and the behavioural response to infection quantified using accelerometry. The student will annotate videos of tagged great tits to identify distinct accelerometry profiles associated with different behaviours. Following this, a behaviour classification model will be built on the video training data (in R) and applied to accelerometry data collected from birds following immune challenge. The student will then examine how the frequency of different behaviours changes with increasing infection exposure. The extent to which the student gets involved in fieldwork will depend on when they start the project and/or level of interest.
Contact Hannah Watson for more details: hannah.watson@biol.lu.se
Â
Suggested reading:
Yu et al. 2024. Flight activity ad effort of breeding pied flycatchers in the wild, revealed with accelerometers and machine learning. Journal of Experimental Biology, 227:jeb247606. https://doi.org/10.1242/jeb.247606
Yu et al. 2023. Accelerometer sampling requirements for animal behaviour classification and estimation for energy expenditure. Animal Biotelemetry, 11:28. https://doi.org/10.1186/s40317-023-00339-w