CabriTrack: Accelerometer data for automated behavioural monitoring of grazing goats

This dataset contains annotated acceleration data of goats grazing outdoors. We used 59 different Creole goats that were equipped with a tri-axial accelerometer of the brand Axivity, model AX3. Two to five animals were set up at the same time inside an experimental field, each animal having an accelerometer located on the horn. The accelerometer was located on the inside of the left horn (for an observer located at the back of the animal). The X axis is along the horn, the Y axis along the back of the animal, and the Z axis perpendicular and on the same plane as the Y axis. Accelerometers recorded the acceleration with a frequency of 25Hz. Animals were monitored with a CCTV camera for subsequent data annotation. Videos were later manually analyzed with the software Boris to label the activity of each animal. Five different behaviors were considered: grazing, ruminating/chewing, displacement, resting, and other. Grazing is when the animal is either with the head down collecting forage or head up chewing. Chewing sequences that lasted less than 5 seconds were labeled as grazing. When chewing sequences lasted more than 5 seconds, they were labeled as ruminating/chewing. Ruminating/chewing thus included chewing sequences of more than 5 seconds, as well as any regular rumination sequences, with the jaw moving. During rumination sequences, the jaw can stop moving for some time. When it stopped for less than 3 seconds, the sequence was still labeled as ruminating. When the sequence stopped for more than 3 seconds, it was labeled as resting. Resting is used for all the sequences where the animal has no activity, either lying on the ground or standing, with the head up or down. Note that alert behavior, with the head of the animal moving rapidly in different directions, was not labeled as resting. Displacement is when the animal explores the field, walking or running. Transitions between postures were also accounted for in displacement, such as when the animal moves from standing to lying or from lying to standing. The dataset is composed of more than 144 hours of annotated behavior from 59 different animals. It contains 12,756,669 lines and 8 columns, each line being an acceleration data point with the metadata. The first column, Time, is the timestamp of the data point. The Animal_id column is the ID of the animal from which the data were acquired. We did not use the national identification number, but instead integers from 1 to 59. The Behaviour column gives the labeled behavior for the sequence, with values in {Displacement, Grazing, Ruminating_Chewing, Other, Resting}. The Behaviour_num column gives the behavior of the sequence as an integer, with the same order used in the previous list (e.g., 1 for Displacement). The columns X, Y, and Z give the acceleration value for the corresponding axis. The Sequence_num is used to identify the behavior sequences. Lines of the dataset with the same Sequence_num are from the same behavior sequence, i.e., they are from consecutive timestamps, from the same animal, and with the same behavior.