Redes neurais artificiais na predição do consumo e produção de leite por vacas F1 Holandês x Zebu

Abstract

The productivity and profitability of ruminant production systems are strongly influenced by dry matter intake (DMI). In the case of dairy cows, dry matter intake plays a crucial role in milk production and quality, as well as overall health and well-being of the animal. Accurately predicting DMI is essential for formulating diets that maximize milk production while ensuring animal welfare. Neural networks have emerged as a viable alternative to traditional statistical models in the field of forecasting methodologies. Thus, the objective of this study was to use neural network methodology to predict the consumption and milk production of F1 Holstein x Zebu cows in semi-arid regions. The dataset used was from dissertations carried out at UNIMONTES, with input variables for neural network construction related to nutrient consumption, digestibility, microbial protein synthesis, animal behavior, and milk composition. The output variables were dry matter intake and milk production. They were used 70% of the data for network training, and 30% for validation. The procedures were carried out using the Neural Network Toolbox in Matlab® 6.5. Two experiments were conducted: Experiment 1 predicted dry matter intake, and Experiment 2 predicted milk production. The neural network was efficient in its predictions, both in terms of similarity to actual consumption and production, as well as low mean squared error. The actual dry matter intake was 15.83 kg/day, while the neural network predicted a consumption of 15.76 kg/day. Milk production was 14.74 kg/day, while the predicted one was 14.62 kg/day. When comparing the real values of dry matter consumption with some prediction equations, both underestimation and overestimation of consumption were observed.