Predicting performance traits in Murrah buffaloes using machine learning: a comparative approach

This study evaluated the comparative performance of nine machine learning (ML) algorithms for predicting 305-day first lactation milk yield (305FLMY) and total milk yield (TMY) in Murrah buffaloes. Data from 657 animals recorded over 24 years (2000-2023) were used, incorporating inputs such as animal details, year of calving, age at first calving (days), peak yield (kg), days to peak yield (DPY), and test day milk yields on the 6^th (TD1), 35^th (TD2), and 65^th (TD3) days. The ML algorithms included Artificial Neural Networks (ANN), Bayesian Regression (BR), Gaussian Process (GP), Gradient Boosting Machines (GBM), Multivariate Adaptive Regression Splines (MARS), Multiple Linear Regression (MLR), Random Forest (RF), Sequential Minimal Optimization Regression (SMOreg), and Support Vector Machines (SVM). Model performance was assessed using R², root mean square error (RMSE), mean absolute error (MAE), Mean absolute percentage error (MAPE), and bias. Among these, the RF model outperformed others for predicting 305FLMY (R² = 78.43%, RMSE = 258.41, MAPE = 9.46%), while SVM provided the best predictive performance for TMY prediction (R² = 71.76%, RMSE = 349.32, MAPE = 276.13). Conversely, ANN demonstrated the weakest performance across both traits. These findings indicated the potential of RF and SVM models for accurate prediction of complex traits in buffalo breeding programs. Future research should aim to enhance model interpretability and computational efficiency for practical on-farm application.