Supervised Machine Learning Techniques for Breeding Value Prediction in Horses: An Example Using Gait Visual Scores.

Area of Science:

• Animal Genetics
• Quantitative Genetics
• Machine Learning in Animal Breeding

Background:

• Gait scores are crucial for genetic evaluation in horses, but subjectivity in phenotypic data can hinder genetic progress.
• Objective measurement of gait traits is essential for accurate genetic selection and improving horse performance.

Purpose of the Study:

• To evaluate the efficacy of machine learning techniques for predicting breeding values (EBV) of five visual gait scores in Campolina horses.
• To compare the performance of artificial neural networks (ANN), random forest regression (RFR), and support vector regression (SVR) against traditional multiple-trait models (MTM).

Main Methods:

• Utilized a dataset of over 5000 phenotypic records and a 14-generation pedigree for 107,951 Campolina horses.
• Estimated variance components and EBVs using a multiple-trait model (MTM).
• Trained ANN, RFR, and SVR models using adjusted phenotypes and fixed effects solutions, with MTM EBVs as the target variable. Validated models using linear regression.

Main Results:

• Machine learning models demonstrated accuracy comparable to MTM, with ANN showing slightly higher accuracy.
• ANN exhibited the highest bias, followed by MTM, while dispersion varied, being highest for ANN and lowest for MTM.
• All tested machine learning models proved to be a feasible alternative for EBV prediction in gait traits.

Conclusions:

• Machine learning presents a viable approach for predicting breeding values for subjective traits like gait scores in horses.
• While accurate, machine learning methods may introduce slight bias and over-dispersion, especially in younger horses, warranting careful consideration in genetic evaluations.