Use of supervised learning for the analysis of weight and degree of finishing of cattle in Precision Livestock

• Grant number: 20/03941-4
• Support Opportunities: Scholarships in Brazil - Post-Doctoral
• Effective date (Start): July 01, 2020
• Effective date (End): October 14, 2021
• Field of knowledge: Physical Sciences and Mathematics - Mathematics - Applied Mathematics
• Principal Investigator: Luis Gustavo Nonato
• Grantee: Adriele Giaretta Biase
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Associated research grant: 13/07375-0 - CeMEAI - Center for Mathematical Sciences Applied to Industry, AP.CEPID

Abstract

Brazilian beef cattle stands out on the international scene for having the largest commercial herd in the world. Improving productive efficiency in Livestock will bring enormous economic and environmental benefits to the country. This improvement can occur through the integration of technologies and the adoption of mathematical tools that allow to evaluate and predict the growth dynamics and body composition of each animal. Predicting the characteristics of animals can increase the possibilities of trading in the physical market, as in the future. Many prediction models take into account only the weight of the animal based on automatic scales. It should be noted that there are major challenges in the integration of a single model that is able to aggregate the degree of finish, climatic data, nutrition, weight and the value of the fluctuation of the arroba over time in a multivariate way in order for producers to negotiate in search the lowest possible risk. Many slaughterhouses offer bonus or discount values related to animals, according to some characteristics such as race, gender, final weight, degree of finish and carcass yield, presence or absence of hump. These are, therefore, the attributes that have the greatest impact at the time of trading. The prediction of weights and degree of finishing of cattle should help in anticipating decisions and provide greater mobility for the negotiation of animals, thus offering an effective strategic planning that improves the productive efficiency of the Livestock sector. Weight predictions also contribute to genetic improvement, since it allows the identification of individuals who grow more efficiently, even under the influence of environmental variables. This project aims to contribute to the state of the art of this important theme in the context of Brazilian agribusiness by fulfilling the following objectives: 1) to predict the weight and the degree of finishing of the carcass of cattle using supervised learning as: Distance-Weighted k-Nearest-Neighbor (DWNN), Multi Layer Perceptron (MLP) and Recurrent Neural Networks, specifically Long Short-Term Memory network (LSTM); 2) evaluate the accuracy and precision of weight prediction at different times in the animals' finishing phase; 3) risk analysis (net profit margin) of the producers, assisting in the optimum economic point of sale of the animals. (AU)

Scientific publications

(References retrieved automatically from Web of Science and SciELO through information on FAPESP grants and their corresponding numbers as mentioned in the publications by the authors)

BIASE, ADRIELE GIARETTA; ALBERTINI, TIAGO ZANETT; DE MELLO, RODRIGO FERNANDES. On supervised learning to model and predict cattle weight in precision livestock breeding. COMPUTERS AND ELECTRONICS IN AGRICULTURE, v. 195, p. 19-pg., 2022-04-01. (20/03941-4)