Classification of nest-building behaviour in non-crated farrowing sows on the basis of accelerometer data

• We detect nest-building behaviour with sensitivity of 87%, specificity of 85%.
• Accuracy of automatic detection is 86%.
• In 7 out of 9 sows at least 5 alarms were generated before farrowing started.

The objective of the study was to test the effectiveness of classification, on the basis of accelerometer data, of nest-building behaviour in farrowing sows that are not confined in crates. The experiment took place in the research farm of University of Veterinary Medicine Vienna, using a herd of 120 Edelschwein sows. Data were collected from 9 sows housed in farrowing pens. The behaviour of 9 sows was video recorded and labelled for a period of 24 h before farrowing, with focus on nest-building activities. Each sow had a specific ear tag with an 3 axis accelerometer sensor mounted on the ear. Acceleration was measured at a frequency of 10 Hz. Out of nine sows under observation, six were assigned to the training set and three as a test set. Classification of nest-building events in the test set using accelerometer data with the generalised linear model indicated sensitivity of 87%, specificity of 85% and accuracy of 86%. The developed technique can be used as part of a Precision Livestock Farming (PLF) automatic monitoring system, where PLF can be defined as management of livestock production using the principles and technology of process engineering. On-farm application of the system would give the possibility to keep sows unconfined until the end of nest-building period. Thus, crating of individual sows could be limited to the first few days after farrowing when the risk of piglet crushing is high. This would improve welfare of sows, without an increase in piglet mortality and without extra labour demand for observation.