A Wearable Software Sensor for Parturition Onset Prediction in Sows

Parturition is a crucial event in the reproductive cycle of sows, and accurate prediction of its onset is essential for optimizing management practices. Traditional methods of prediction often rely on visual observations and manual monitoring, which may not always be precise and timely. Therefore, in this study a wearable non-invasive software-sensor was used to predict the onset of sow parturition. The sensor system is designed to continuously monitor activity, body heat flux, and skin temperature. The system incorporates an adaptive subject-specific model to predict the time of parturition. This innovative approach will aid the optimization of sow management around parturition, reducing the risk of birth complications and piglet losses. A total of seven healthy TN70 sows are used for the purpose of this study (parity 2.3 ± 1.6). The sensor system is attached directly onto the skin, at the base of the ear, ± 2 days prior to the expected parturition date. The measurements are stopped 24h post-parturition. Rectal temperature was measured twice daily to validate the sensor’s temperature measurements. Parturition duration, and sow activity are visually monitored using continuous video recordings as a gold standard. Cox proportional hazards and one-class support vector machine models is trained to predict parturition based on different prediction horizons, namely, 2, 4, 6, and 8 hours before the exact time of parturition. The results from both the Cox Proportional Hazards (CPH) model and the One-Class Support Vector Machine (OCSVM) model demonstrate promising potential for predicting the onset of parturition events based on the measured predictor variables obtained from the proposed wearable sensor system. The CPH model achieved a high accuracy of 86% in predicting parturition onset time within the critical time window of 4 to 6.5 hours before the observed events.