Towards qualitative and quantitative prediction and detection of parturition onset in sows using light barriers

•A single light barrier can be used for the prediction and detection of parturition.•A validated classification performance close to 90% was achieved.•Approaches for measuring the parturition related activity increase were compared.•A method for a quantitative prediction of the duration until parturition was tested.

Piglet mortality can be a large economic and animal welfare issue in breeding facilities. A system that predicts the parturition can help the breeder in economically organising staff assignments in order to achieve an optimal workload levelling. In the current study, light barriers at the head and torso region of a sow were used to measure and classify the activity increase of 34 sows related to their near parturition. Based on this data, 4 different activity frequency and activity duration based qualitative predictors for the near onset of parturition were developed retrospectively, utilising cumulative sum techniques and a global threshold approach. The threshold optimisation for the qualitative prediction was performed using a random set of 17 sows and validated with the remaining sows. The best performing qualitative prediction yielded a validated sensitivity of 88% at a precision of 88%. This prediction generated parturition alerts with a 25th percentile of 13 h and a 75th percentile of 20 h before the parturition started. Based on this indicator, a quantitative prediction of the time remaining until the onset of parturition could be developed. This prediction exhibited a mean prediction error of 0.5 h ± 2.6 h (SD) for 88% of the sows over a period of 13–24 h before the onset of parturition. At the same time 12% of the predictions were unusable with a mean prediction error of 12.5 h ± 6.9 h (SD). In addition, a method for detecting the parturition onset with an accuracy of ±4 h, a sensitivity of 88% and a precision of 97% for the head sensor could be obtained. With data from the torso sensor, the performance of the various indicators was generally lower and optimality was achieved with different thresholds. The present study follows other studies showing the general detectability of the parturition related increase in activity using video, light barriers and ultrasonic distance sensors. It is also closely based on earlier studies using accelerometers for individual qualitative parturition detection, with the explicit intent to reproduce these results using light barriers.