Predicting carcass tissue composition in Blackbelly sheep using ultrasound measurements and machine learning methods

Area of Science:

• Animal Science
• Agricultural Engineering
• Data Science

Background:

• Accurate prediction of carcass tissue composition is crucial for livestock management and breeding.
• In vivo ultrasound offers a non-invasive method for assessing animal growth and carcass traits.

Purpose of the Study:

• To evaluate machine learning models for predicting Blackbelly sheep carcass tissue composition.
• To identify the most effective model for estimating total carcass bone, fat, and muscle.

Main Methods:

• Utilized in vivo ultrasound measurements from Blackbelly sheep.
• Applied and compared four machine learning models: decision trees, random forests, support vector machines, and multi-layer perceptrons.
• Predicted total carcass bone (TCB), total carcass fat (TCF), and total carcass muscle (TCM).

Main Results:

• Random forests outperformed other models in predicting TCB, TCF, and TCM.
• Random forests achieved R-squared values of 0.67, 0.69, and 0.76 for TCB, TCF, and TCM, respectively.
• The study reported low mean squared error (MSE) and mean absolute error (MAE) for the random forest model.

Conclusions:

• Machine learning models, particularly random forests, can reliably predict carcass tissue composition from in vivo ultrasound data.
• In vivo ultrasound combined with machine learning provides a viable tool for non-invasive carcass composition assessment in sheep.
• These findings support the use of predictive modeling in livestock evaluation for improved efficiency and accuracy.