MACHINE LEARNING APPLICATIONS FOR MONITORING HEAT STRESS IN LIVESTOCK
Heat stress is harmful to the health and productivity of livestock. Several models have been proposed for assessing heat stress, but these models assume a particular relationship, chosen by the researcher, between environment conditions and a physiological measure of heat stress. These assumptions may not accurately represent the true, underlying relationship. To account for realistic relationships, we employed machine learning algorithms to (1) rank the effect of environmental heat stressors (air temperature, solar radiation, relative humidity and wind speed) on physiological responses (skin temperature, core, body temperature and respiration rate) of dairy cows, and (2) predict core, skin, and hair-coat temperatures of piglets. The advantage of using machine learning algorithms is that they are data-driven procedures that have greater expressive power than previous modeling procedures considered (mechanistic and linear models). This thesis is organized such that Chapter 2 demonstrates an application of machine learning algorithms to predict physiological responses (skin temperature, core-body temperature, and respiration rates) of dairy cows from environmental heat stressors (air temperature, relative humidity, solar radiation, and wind speed) as well as their interaction terms and rank the effect of these on each physiological response. Chapter 3 demonstrates an application of machine learning algorithms to predict physiological responses of piglets from environmental heat stressors for piglets. Chapter 2 demonstrates that neural networks consistently produced the lowest root mean square error, RMSE, in predicting skin temperature, core-body temperature and respiration rate of dairy cows. The RMSE for skin temperature was 0.38 °C; for core-body temperature was 0.41 °C; and for respiration rate was 12 respirations per minute. Ranking of environmental heat stressors showed that air temperature has the largest effect on each physiological response, followed by solar radiation, and thirdly by the interaction of air temperature and relative humidity. Wind speed and relative humidity were inconsequential heat stressors. Chapter 3 demonstrates that neural networks, gradient boosted machines, and random forests were the best algorithms, based on the lowest mean squared error on the testing dataset, to predict rectal, skin-surface, and hair coat-surface temperatures, respectively. This supports the use of machine learning algorithms to predict the physiological temperatures of piglets.
Gebremedhin, Kifle G.
De Sa, Christopher Matthew
Biological and Environmental Engineering
M.S., Biological and Environmental Engineering
Master of Science
dissertation or thesis