A Hybrid CUSUM approach to identify residency and transition periods for animal movement with an application to housed dairy cows

Over recent years there has been a significant advance in tracking technology which has allowed animal movement data to be collected with greater accuracy and in larger quantities. As a result, there has been a need for developing techniques to analyse this data and gather information such as bouts of foraging behaviour in wild animals, understanding how an animal’s movements relate to the resources within an environment or to provide potential indicators of animal welfare. The aim of this report is to develop an automated behavioural classification for residency and transition periods based on two-dimensional position data via a Hybrid CUSUM method. It is hoped that by detecting the amount of movement of an individual over a selected period of time and comparing it with expected results of healthy individuals will lead to an indication of welfare although this CUSUM method can also be used for other purposes such as gathering information about foraging patterns. The versatility of this method is that it can be applied to any animal that exhibits mostly residency and transitory behaviour. In order to automatically identify the residency and transition periods, a Hybrid CUSUM method has been developed and has been applied to a data set involving housed dairy cows in the hope to identify differences in the typical movements of lame cows against the typical movements of non-lame cows. The Hybrid CUSUM has the novelty that the standard deviation is predefined to correspond to the expected deviation of an animal whilst in a state of residency, whilst the mean is calculated directly from the data. It also has the novelty that a second algorithm is initiated once the system is in a state of out-of-control to detect when the system is back in-control and then the CUSUM will restart around a new mean value. The outcome of this report is the ability to automatically identify when cows are resident within three distinct areas of the barn (resting area, feeding area and milking area) and to imply the types of transitions between areas of the barn. In earlier work, attempts to classify the behavioural state of cows using the same data set have been achieved using accelerometer data, whereas this report uses position data to identify residency periods within a particular zone of the barn and infer types of transitions between zones such as resting to feeding or transitions within the same zone such as feeding to feeding. It is found that non-lame cows tend to move around more and spend longer total time within the feeding area than lame cows.