A machine learning system to evaluate physiological parameters and heat stress for sows in gestation crates

Heat stress can easily affect the sow production performance to make huge financial loss. Although many models have been developed to evaluate heat stress, most of them are built for humans, cows, or young pigs, which could not well used in sow. In this study, the extreme gradient boosting (XGBoost) algorithm was specially developed to predict the physiological parameters of sows in gestation crates, including skin temperature, rectal temperature, and respiration rate, which play crucial roles in reflecting heat stress. The best physiological parameter prediction model was used to evaluate the heat stress of the sow and built the warning system. Datasets (1029) were collected from a commercial pig farm, which included environmental parameters of temperature, relative humidity, and air velocity inside the sow house (used as input data), and physiological parameters of sows (used as output data). The results showed that the model to predict skin temperature (skin temperature model of sows, STMS) got best performance, and the XGBoost algorithm had advantages in dealing with nonlinear problems achieving a significant improvement over the linear algorithm. Additionally, a physiological parameters and heat stress assessment system for sow housing was developed by integrating a heat stress threshold determined using STMS with LabVIEW, offering both a new technological solution and valuable insights.