Comparison of Machine Learning Algorithms for Prediction of Lifetime Milk Yield in Gir Cattle (Bos indicus)

Abstract

Lifetime milk yield (LMY) is an economically important trait in dairy cattle; however, its late expression limits early selection of genetically  superior animals. The present study compared the predictive performance of Multiple Linear Regression (MLR), Artificial Neural Network  (ANN), Random Forest (RF), Support Vector Regression (SVR), and Extreme Gradient Boosting (XGBoost) models for early prediction of  LMY in Gir cattle using routinely recorded first-lactation performance traits. Historical production records of 174 Gir cows maintained  at the Cattle Breeding Farm, Kamdhenu University, Junagadh during 1986 to 2010 were used. Predictor variables included calf birth  weight, test-day milk yields, first lactation length, first lactation peak milk yield, first lactation standard milk yield, first lactation total  milk yield, season, period and year of first calving, and first calf sex. The dataset was randomly divided into training (80%) and testing  (20%) subsets. Model performance was evaluated using root mean square error (RMSE), mean absolute error (MAE), coefficient of  determination (R²), and mean absolute percentage error (MAPE). Among the evaluated models, XGBoost exhibited the best predictive  performance (RMSE = 3713.95 L, MAE = 2679.89 L, MAPE = 10.43%, R² = 0.3865), followed by ANN and RF. Feature importance analysis  identified first-lactation production traits as the major determinants of lifetime milk yield. These findings demonstrate that XGBoost is an  effective tool for early prediction of lifetime milk yield and can support selection decisions in organized Gir cattle breeding programmes