Latent variable models in the understanding of animal monitoring data

This paper looks at techniques in the field of machine learning that can be employed to aid the interpretation of intensively gathered sen-sor data from domestic livestock. Given the high levels of reliability afforded through improved battery technology and progressively more powerful small computing devices, condition monitoring on such scales has become widespread but at the expense of the understanding of the relation to the welfare condition that underlies the quantities being measured. Latent class models offer a means of postulating the existence of an abstraction or category label for a given set of observations. In this chapter we look at the additional understanding that 3 progressively sophisticated models can offer in the interpretation of a set of GIS data gathered from a herd of 15 beef cows. We conclude with a review of practical applications where these models may assist understanding of the potentially complex behavioural relationships between individuals and groups of animals.