UTILIZAÇÃO DE REDES NEURAIS ARTIFICIAIS NA PREDIÇÃO DE ESTRESSE EM OVELHAS PRENHAS

Abstract

This work aimed to evaluate the thermal stress during pregnancy and maternity in sheep farming using RNA, based on environmental and physiological thermal variables of pregnant ewes. The research was carried out at the Centro Tecnológico de Ovinos (CTO) of Universidade Anhanguera Uniderp using 30 matrices housed in a common paddock during the pre and post pregnancy phases, where three stress indicators were classified according to the respiratory rate (Light, Moderate and High). To this end, a Perceptron Multilayer RNA was implemented with an input layer, a hidden layer (with seven neurons) and an output layer, with the function of hyperbolic tangent activation and softmax. Ambient air temperature, relative humidity, and skin, body, wool, head, rectal and respiratory rate temperatures were considered to be input variables. Animal stress was considered as the output variable. The results can be considered satisfactory, since the ANN presented a cross entropy error of 2.337 and incorrect predictions of 1.1% in predicting the level of thermal stress of the sheep.