High accuracy at low frequency: detailed behavioural classification from accelerometer data

Abstract

Accelerometers are a valuable tool for studying animal behaviour and physiology where direct observation is unfeasible. However, giving biological meaning to multivariate acceleration data is challenging. Here, we describe a method that reliably classifies a large number of behaviours using tri-axial accelerometer data collected at the low sampling frequency of 1 Hz, using the dingo (Canis dingo) as an example. We used out-of-sample validation to compare the predictive performance of four commonly used classification models (random forest, k-nearest neighbour, support vector machine, and naïve Bayes). We tested the importance of predictor variable selection and moving window size for the classification of each behaviour and overall model performance. Random forests produced the highest out-of-sample classification accuracy, with our best-performing model predicting 14 behaviours with a mean accuracy of 87%. We also investigated the relationship between overall dynamic body acceleration (ODBA) and the activity level of each behaviour, given the increasing use of ODBA in ecophysiology as a proxy for energy expenditure. ODBA values for our four 'high activity' behaviours were significantly greater than all other behaviours, with an overall positive trend between ODBA and intensity of movement. We show that a random forest model of relatively low complexity can mitigate some major challenges associated with establishing meaningful ecological conclusions from acceleration data. Our approach has broad applicability to free-ranging terrestrial quadrupeds of comparable size. Our use of a low sampling frequency shows potential for deploying accelerometers over extended time periods, enabling the capture of invaluable behavioural and physiological data across different ontogenies.

Optimisation of energetic and reproductive gains explains behavioural responses to environmental variation across seasons and years

“…Individual squirrels were captured on defended territories, weighed, assessed for reproductive condition, and fitted with an accelerometer (models Axy2/Axy3, 4 g [1.7% of body mass], Technosmart Europe) in collar form, either ventrally mounted on its own ( n = 128) or dorsally‐mounted in combination with a ventrally mounted VHF radio transmitter ( n = 361, model PD‐2C, 4 g [1.7% of body mass], Holohil Systems Limited, Carp, ON, Canada; see Studd, Landry‐Cuerrier, et al, 2019 for collar design). All accelerometers recorded acceleration between ± 8 g forces at a sampling rate of 1 Hz and temperature at a rate of 0.1 Hz, frequencies that have been shown to capture broad‐scale behaviour of small animals with high accuracy, allowing for long‐duration recordings (Tatler et al, 2018; Studd, Boudreau, et al, 2019). Squirrels were released at site of capture and remained free‐ranging until recaptured for collar removal (3–103 days later).…”

“…Meanwhile, consumer resource theory allows for behaviour to affect the consumption of resources, but treats energy expenditure as behaviourally independent (Yodzis & Innes, 1992; Post et al, 2000). The methodological constraint requiring direct observation of behaviour has now largely been eliminated by recent advances in biologging technologies which offer effective methods for continually recording fine‐scale behavioural variation (Kays et al, 2015) over long durations (Williams et al, 2016; Tatler et al, 2018; Studd, Boudreau, et al, 2019). Accordingly, we focus the next two paragraphs on describing an empirical approach for categorising behavioural variation and a conceptual approach to relating these behavioural categories to their energetic and ecological consequences.…”

Machine learning goes wild: Using data from captive individuals to infer wildlife behaviour

“…Our aim was to find a window size with a high mean performance that is small enough for generating sufficient data. Unfortunately, reducing the window size was found to negatively impact model performance (Tatler et al, 2018). Performance seemed to be at its maximum at window size 79.…”

“…Finally, we analysed the ODBA, an indicator of body movement (Wilson et al, 2006) that was shown to correspond well with the activity level of specific behaviours (Tatler et al, 2018).…”

“…The Random Forest (RF) and the Support Vector Machine (SVM) are popular approaches to infer animal behaviour from acceleration data and have yielded good results under training conditions (Nathan et al, 2012;Tatler et al, 2018). However, we are not aware of any study successfully transferring a behaviour classification model trained on captive individuals to wild individuals.…”