Advanced search
Start date
Betweenand

Prediction of thermal comfort level for dairy cattle: method based on machine learning modelling using climate chamber experiment data

Grant number: 19/26828-1
Support type:Scholarships in Brazil - Master
Effective date (Start): July 01, 2020
Effective date (End): January 31, 2022
Field of knowledge:Agronomical Sciences - Agricultural Engineering
Principal Investigator:Rafael Vieira de Sousa
Grantee:Alex Vinicius da Silva Rodrigues
Home Institution: Faculdade de Zootecnia e Engenharia de Alimentos (FZEA). Universidade de São Paulo (USP). Pirassununga , SP, Brazil

Abstract

Recent studies in animal production have investigated technologies and computational models for predicting the level of thermal comfort through noninvasive and automatic measurements. To contribute to this theme, the project aims to construct and test computational models to predict the thermal comfort level of dairy cattle using a database constructed by climate chamber experiment. The experiment will be conducted for 25 days with 12 Holstein heifers randomly assigned to two groups: stress group and control group. The animals in the stress group will be allocated to the climate chamber (in a Tie Stall system) and exposed to two heat waves. The control group will remain in individual pens partially covered in room temperature containment. During the period of the experiment, besides the meteorological data of the facilities, will be collected 5 times a day (6, 10, 14, 18 and 22 hours) data of rectal temperature, respiratory rate, sweating rate and body surface temperature in different areas of the body (eye, forehead, rib and flank) by infrared thermography. For the modeling step, different Machine Learning based algorithms (artificial neural networks, support vector machine, random forest and k-nearest neighbors) will be evaluated using different combinations of prediction inputs or thermal comfort level attribute classification. The tests to compare the prediction models will use linear regression (correlation coefficient and residual) and error (mean percentage error and mean square error) as an analysis metric. To compare the classification models will be used as metric the parameters obtained from the confusion matrix generated (accuracy, precision, sensitivity and F1 score). Through the project results we seek to highlight not only the best computational techniques for modelling, but also to highlight the potential of using a climate chamber experiment in order to create a more homogeneous and faster database in relation to techniques based on outdoor experiment. (AU)