Unravelling domestic cat behaviour using accelerometers and machine learning : a thesis presented in partial fulfilment of the requirements for the degree of Doctor of Philosophy in Animal Science at Massey University, Palmerston North, New Zealand

Abstract

It is estimated that there are at least 445 million companion cats worldwide. Despite this large number, little is known about how environmental factors (e.g., weather, presence of children, dogs or other cats) are associated with domestic feline behaviour and welfare. Traditional behavioural observation methods are limited by short observational windows, high labour demands, and subjective nature of observations, making them unsuitable for in-depth, longitudinal studies. This thesis leverages recent advances in accelerometer technology and machine learning to overcome these limitations to quantitatively analyse continuous domestic cat behaviour, providing new insights into the influence of environmental factors on their behaviour. In the first phase of the thesis (Chapter 3), machine learning models were trained to classify cat behaviours from acceleration data. The models achieved a minimum accuracy of 70%, which improved as the number of behavioural classes was reduced, and with random forest models generalising better to new data than self-organising maps. Models for harness-mounted devices (accuracies 77% - 83%) performed slightly better than those for collar-mounted devices. The model for the collar-mounted device that classified eight different behaviours (active, eating, grooming, littering, lying, scratching, sitting, standing), performed adequately (accuracy 73%), and was considered the most practical due to cats being more likely to wear a collar than a harness. The second and third phases of this thesis investigated how environmental factors influenced cat behaviour, both in semi-outdoor research cats (Chapter 4) and in pet cats within a home environment (Chapter 5). Weather variables, particularly daylength and the temperature humidity wind index, were significantly associated with domestic cat behaviour. Seasonal changes in grooming and scratching were closely linked to the natural hair growth cycle. Behaviours were influenced by individual differences and disruptions, such as the return of a cat to the group after an absence, affecting their behaviour. Among pet cats, those with outdoor access showed behavioural adaptations to seasonal weather changes, seemingly prioritising thermal comfort. Multi-cat households and those with at least one child present were associated with an increase in alertness-related behaviours of increasing sitting with decreasing lying. By leveraging machine learning and accelerometer data, this research advances current methodologies in animal behaviour studies, providing an approach for continuous, minimally invasive animal behaviour and welfare assessment. This thesis contributed to bridging the gap in our understanding of how environmental factors can shape animal behaviour and welfare and highlighted the importance of taking interindividual differences and environmental factors into account when assessing domestic cat behaviour.