Comparative Evaluation of Wood, Wilmink, Guo-Swalve, and Random Regression Models for Describing Lactation Curves in Dairy Buffalo

This study compared four lactation curve models—Wood, Wilmink, Guo–Swalve, and Random Regression Model (RRM) using 9,065 lactations from Nili-Ravi buffaloes, each with 43 test-day milk yield (TDMY) records collected between 2001 and 2024. Models were evaluated based on goodness-of-fit (AIC, BIC), predictive accuracy (MSPE, RMSE), and biological parameters (peak yield, persistency, time to peak). Among the four models, RRM provided the best overall fit, while Wilmink demonstrated the highest prediction accuracy, characterised by low error and bias. The Guo–Swalve model showed intermediate performance, while the Wood model underperformed in both fit and prediction. RRM closely estimated peak yield and time to peak (≈ 45 days in milk), and exhibited the highest persistency, suggesting value for genetic evaluation. Wilmink maintained stable prediction error across lactation stages and offered an effective balance of simplicity and precision. These results underscore that model selection should be goal-specific: RRM is best for genetic evaluations and capturing biological variation, while Wilmink is optimal for operational forecasting and management.