Detection of rumination in cattle using an accelerometer ear-tag: A comparison of analytical methods and individual animal and generic models

On-animal sensors are revolutionising livestock farming by automating individual animal monitoring. Rumination is closely linked to cow health and physiology and is a perfect candidate for monitoring using on-animal sensors. The purpose of the current study was to determine if a tri-axial accelerometer ear-tag could be used to detect rumination behaviour in eight multiparous Angus crossbreed cows housed in a semi-enclosed barn. Different machine learning algorithms (classification and regression tree, random forest, linear discriminant analysis, quadratic discriminant analysis) and epoch lengths (1 s, 5 s, 10 s, 30 s, 60 s, 90 s, mixed) were tested for ability to predict rumination. Two approaches were taken to develop rumination prediction models: 1) a generic rumination prediction model developed from the data of all eight cows (generic model) and 2) an individual rumination prediction model (individual model). The most accurate generic model utilised a classification and regression tree with a mixed epoch (accuracy = 86.2%, sensitivity = 75.3%, specificity = 92.5%). Accuracy, sensitivity, and specificity improved when using the proposed ‘individual model’ with an average accuracy of 98.4%. The current study details how rumination can be modelled using accelerometer ear-tag technology, and provides insight as to how the described algorithms may be integrated in commercial contexts.