Genomic Selection for Cashmere Traits in Inner Mongolian Cashmere Goats Using Random Forest, Gradient Boosting Decision Tree, Extreme Gradient Boosting and Light Gradient Boosting Machine Methods.

In recent years, Machine Learning (ML) has garnered increasing attention for its applications in genomic prediction. ML effectively processes high-dimensional genomic data and establishes nonlinear models. Compared to traditional Genomic Selection (GS) methods, ML algorithms enhance computational efficiency and offer higher prediction accuracy. Therefore, this study strives to achieve the optimal machine learning algorithm for genome-wide selection of cashmere traits in Inner Mongolian cashmere goats. This study compared the genomic prediction accuracy of cashmere traits using four machine learning algorithms-Random Forest (RF), Extreme Gradient Boosting Tree (XGBoost), Gradient Boosting Decision Tree (GBDT), and LightGBM-based on genotype data and cashmere trait phenotypic data from 2299 Inner Mongolian cashmere goats. The results showed that after parameter optimization, LightGBM achieved the highest selection accuracy for fiber length (56.4%), RF achieved the highest selection accuracy for cashmere production (35.2%), and GBDT achieved the highest selection accuracy for cashmere diameter (40.4%), compared with GBLUP, the accuracy improved by 0.8-2.7%. Among the three traits, XGBoost exhibited the lowest prediction accuracy, at 0.541, 0.309, and 0.387. Additionally, following parameter optimization, the prediction accuracy of the four machine learning methods for cashmere fineness, cashmere yield, and fiber length improved by an average of 2.9%, 2.7%, and 3.8%, respectively. The mean squared error (MSE) and mean absolute error (MAE) for all machine learning methods also decreased, indicating that hyperparameter tuning can enhance prediction accuracy in ML algorithms.