MEAT TENDERNESS PREDICTION USING MACHINE LEARNING: DISTRIBUTION APPLIED TO VARIABLES

Resumo

Beef tenderness is an attribute of great prestige for consumers. Several factors, such as genetics, food,
and environment, influence meat tenderness, which can be objectively evaluated after the animal is slaughtered.

Typically, sensitivity measurement involves mechanical testing, in which the shear force required to break mus-
cle fibers is quantified. This study aims to validate a more comprehensive and robust database, incorporating

hyperspectral imaging, to help predict meat tenderness non-destructively. In this new approach, computational

techniques, such as machine learning with the use of artificial neural networks, are employed to explore the depen-
dence of transmitted variables without the need to control the response. Hyperspectral images provide information

about specific wavelengths, allowing for a more detailed analysis of the sample. To assess tenderness parameters,

measures such as pH, sample color, hot carcass weight, ribeye area, breed, and sex, as well as hyperspectral im-
ages, along with fillet shear force. The objective is to identify the relationship between these variables and the

shear force required to break the muscle fibers. The use of normal and gamma mathematical distributions as tools
for more comprehensive training in Machine Learning was used. The analysis showed that the statistical model
adequately adjusted the data, confirming the reliability and accuracy of the forecasts obtained. This additional
validation reinforces the robustness of the proposed model, which can handle the variability of beef tenderness
data. With these new arrangements, it was possible to study the behavior of these distributions through validation
from a robust database in which Random Forest algorithms were implemented in Machine Learning and Neural
Network. Based on the data presented, a Coefficient of Determination (R2) of 0.4494 was obtained, demonstrating
the effectiveness of the Machine Learning model in predicting the shear force in relation to the values obtained in
the mechanical tests. This innovative approach, using hyperspectral imaging in conjunction with machine learning
techniques, provides a broader and more robust database for predicting beef tenderness. These scientific advances
have the potential to improve end-product quality and meet consumer expectations for beef tenderness.