Morphological characterization of the Polatli sheep in terms of live weight using data mining algorithms.

Area of Science:

• Animal Science
• Agricultural Engineering
• Data Science

Background:

• Accurate live weight (LW) estimation is crucial for livestock management and breeding programs.
• Morphological characteristics offer a non-invasive method for assessing animal growth and condition.

Purpose of the Study:

• To estimate the live weight (LW) of Polatli sheep using body measurements and machine learning algorithms.
• To identify significant independent variables for live weight estimation.
• To compare the predictive performance of Classification and Regression Tree (C&RT), Chi-square Automatic Interaction Detector (CHAID), and Multivariate Adaptive Regression Splines (MARS) algorithms.

Main Methods:

• Utilized body measurements (withers height, rump height, body length, chest depth, chest width, chest girth, cannon bone circumference), age, and sex as independent variables.
• Applied C&RT, CHAID, and MARS algorithms for live weight prediction.
• Employed 10-fold cross-validation and specific node settings for tree-based algorithms.

Main Results:

• All algorithms showed strong correlations between body measurements and live weight (p < 0.05).
• The MARS algorithm achieved the best prediction performance across multiple metrics (RMSE, SDR, MAPE, Adj-Rsq, AIC).
• C&RT outperformed CHAID among the tree-based algorithms.

Conclusions:

• C&RT and MARS algorithms are reliable for morphological characterization and identifying indirect criteria for live weight.
• These algorithms can aid in forming elite herds and optimizing breeding programs.
• Body measurements are effective predictors of live weight in Polatli sheep.