Observing animals in the wild often poses extreme challenges, but animal-borne accelerometers are increasingly revealing unobservable behaviours. Automated machine learning streamlines behaviour identification from the substantial datasets generated during multi-animal, long-term studies, however, the accuracy of such models depends on the qualities of the training data. We examined how data processing influenced the predictive accuracy of random forest (RF) models, leveraging the easily observed domestic cat (Felis catus) as a model organism for terrestrial mammalian behaviours.

Nine indoor domestic cats were equipped with collar-mounted tri-axial accelerometers, and behaviours were recorded alongside video footage. From this calibrated data, eight datasets were derived with; (i) additional descriptive variables; (ii) altered frequencies of acceleration data (40 Hz vs. a mean over 1 second); and (iii) standardised durations of different behaviours. These training datasets were used to generate RF models which were validated against calibrated cat behaviours before identifying behaviours of five free-ranging tag-equipped cats. These predictions were compared to those identified manually to validate the accuracy of the RF models for free-ranging animal behaviours.

RF models accurately predicted the behaviours of indoor domestic cats (F-measure up to 0.96) with discernible improvements observed with post-data-collection processing. Additional variables, standardized durations of behaviours, and higher recording frequencies improved model accuracy. However, prediction accuracy varied with different behaviours, where high-frequency models excelled in identifying fast paced behaviours (e.g. locomotion), while lower frequency models (1 Hz) more accurately identified slower, aperiodic behaviours such as grooming and feeding, particularly when examining free-ranging cat behaviours.

While RF modelling offered a robust means of behaviour identification from accelerometer data, field validations were important to validate model accuracy for free-ranging individuals. Future studies may benefit from employing similar data processing methods that enhance RF behaviour identification accuracy, with extensive advantages for investigations into ecology, welfare, and management of wild animals.

Indoor cats were fitted with neck collars to which tri-axial accelerometers recording at 40 Hz were affixed. Accelerometer data was synchronised with video footage of the cats and distinct behaviours were labelled (‘rest’, ‘walk’, ‘trot’, ‘run’, ‘collar shake’, ‘feed’, and ‘groom’) using bespoke software DDMT (Wildbyte technologies, http://wildbytetechnologies.com/software.html). Transitions between behaviours were not included in any behaviour sample.

Details

Nine adult domestic cats (4 females, 5 males; aged 6 months – 8 years) housed at Mid Antrim Animal Sanctuary, Antrim, Northern Ireland in rooms (2 m x 3 m) were studied in June and July 2017. Cats were free to move to an enclosed outside area (2 x 2 m). All individuals were either neutered or spayed and were certified as healthy by a veterinarian prior to participation in the study.

Cats were fitted with quick release collars (Breakaway buckle collar, Rogz Ltd. 2002/030628/07) to which a tri-axial accelerometer (‘Daily Diary’, Wilson et al. 2008) recording at 40 Hz was attached. The total weight of the collar and logger was 25g (less than 1% of the cats’ body weight). Daily Diary loggers were fitted under the chin of the cat in line with the lateral (sway, X), vertical (heave, Y), and sagittal (surge, Z) body axes. Whilst wearing the collars, cats were filmed using a Sony Alpha a58 DSLR camera (Sony, Latin America, Inc.) for 15 minutes during the morning while a second researcher encouraged the cat to undertake different behaviours, such as running after a toy, or provided food to observe feeding behaviours. In addition, naturally occurring behaviours were observed, such as walking, trotting, resting, grooming, and shaking the collar. These behaviours were selected as they accounted for much of the cat’s (and other wild equivalent predator’s) daily behaviours, are of ecological significance, and the repertoire can be indicative of welfare. Each cat was filmed for 15 minutes to record the different behaviours it undertook.

Accelerometer data and video synchronisation

Accelerometer data and video footage from the indoor cats were synchronised using the timestamp of the data and video. To guard against any potential inaccuracies of their internal clocks, during the video, the collar was shaken up and down by the observers to create a distinct marking point in the accelerometer data that could be synchronised with the camera timestamp on the recording. Any offset that was required between the camera and accelerometer was noted and added to the accelerometer data once the data was downloaded and loaded into DDMT software. Distinct behaviours that lasted at least two seconds were selected on the video, and identified in the accelerometer data via the corrected timestamp. Transitions between behaviours were not included in any behaviour sample. The data was also examined to confirm it contained likely values for this behaviour (e.g. resting did not contain extremely high accelerometer values) to ensure behaviours were not incorrectly classified. DDMT is a specialised accelerometer handing software that allows various data manipulations to be done, including “labelling” of behaviours that can then be extracted individually. This was conducted for all distinct identifiable behaviours within the video footage and the accelerometer samples extracted.

Ethical Approval

This project was given ethical approval by committee at Queen’s University Belfast (QUB-BS-AREC-19-005).