Evaluation of Selected Machine Learning Algorithms to Predict Longevity in Iranian Holstein Dairy Cattle

The increasing need for sustainable dairy production has emphasized the importance of improving longevity in dairy cattle, which directly impacts farm profitability and herd efficiency. Traditional statistical models often fall short in capturing the complex, nonlinear factors influencing this trait, highlighting the potential of Machine Learning (ML) approaches. This study followed a retrospective observational design within a predictive regression modeling framework. This study evaluated the predictive performance of five models: Gradient Boosting Machine (GBM), Random Forest (RF), Decision Tree (DT), Support Vector Machine (SVM), and a conventional Linear Model (LM) to estimate the longevity of Iranian Holstein dairy cattle. The dataset comprised 422,751 records from 12,807 sires across 5,417 herds, including production, reproductive, and geographic data over ten years. Longevity was defined as the interval from first calving to removal from the herd, and models were trained and tested using 10-fold cross-validation under a regression framework. Among the tested models, SVM achieved the highest predictive performance with a coefficient of determination (R²) of 0.989 and the lowest Root Mean Squared Error (RMSE) of 2.51. RF and GBM also performed well (R² = 0.983 and 0.979, respectively). The most influential predictors across models were the number of calvings, average calving interval, and first calving year, highlighting the importance of reproductive history. While ML models showed superior predictive power, they also presented challenges such as higher computational demands and potential overfitting if not properly validated. The findings suggest that advanced ML models, particularly SVM and RF, offer valuable tools for improving decision-making and culling strategies in dairy herd management. Future research should validate these models on independent datasets and explore their integration into practical selection and monitoring systems.