Comparison of the data mining and machine learning algorithms for predicting the final body weight for Romane sheep...

Area of Science:

• Animal Science
• Machine Learning
• Genetics

Background:

• Accurate prediction of final body weight is crucial for selecting superior Romane sheep for reproduction and meat yield.
• Traditional selection methods can be enhanced by utilizing data mining and artificial intelligence algorithms.

Purpose of the Study:

• To predict the final body weight of Romane sheep using various machine learning algorithms.
• To identify key factors influencing final body weight for indirect selection strategies.

Main Methods:

• Employed Classification and Regression Tree (CART), Multivariate Adaptive Regression Splines (MARS), and Support Vector Machine Regression (SVR) algorithms.
• Trained and tested models on a dataset of 393 Romane sheep, utilizing features such as birth weight, sex, and weaning data.
• Evaluated model performance on training (80%) and testing (20%) sets.

Main Results:

• The CART model demonstrated the best predictive performance for final body weight in both training and testing datasets.
• Identified sex, suckling weight, weaning weight, age at weaning, and age at final weight as significant predictors.
• Proposed criteria for genetic improvement: female Romane sheep over 142 days old with weaning weight exceeding 28 kg.

Conclusions:

• The CART algorithm is a valuable tool for predicting Romane sheep final body weight and facilitating indirect selection.
• The findings can aid in establishing breed standards and identifying superior animals for enhanced meat production in France.
• Integrating machine learning models offers a promising approach for genetic improvement in sheep farming.