Determining the Best Predictive Function for mathematically describing the Lactation Curve in Murcia Goats in Iran

This study compared Wood, Wilmink, Pollott, and Dijkstra functions to fit lactation curves using 38,030 test-day (TD) milk records from first-parity Murcia goats (2018-2024). Two scenarios were analyzed: one included all animals (TD greater than 3) using Wood, Wilmink, and Pollott, while the other utilized complete records (8 TD) and all four functions. In the first scenario, Pollott demonstrated the highest percentage of typical lactation curves (81.2%) among the tested functions, followed by Wood (71.4%) and Wilmink (61.5%). Additionally, Pollott exhibited the lowest correlation among its parameters, reducing the potential for multicollinearity and overfitting. Although Pollott performed well, it had higher Akaike's Information Criterion (AIC) and Bayesian Information Criterion (BIC) values compared to the others. However, the corrected AIC (AICc) value for Pollott was lower, suggesting superior performance in smaller datasets. In contrast, Wood consistently showed the lowest AIC and BIC values across both scenarios, indicating that it was the most robust in terms of function fit. Despite this, a comparison of the predicted lactation curves revealed that Pollott more accurately predicted daily milk production, especially during the middle of lactation, although it slightly overpredicted milk yield at the start of lactation. Furthermore, the analysis of parameter correlations and variance inflation factors (VIF) highlighted that Pollott avoided significant multicollinearity issues, which were more prominent in other functions. Therefore, Pollott, due to its flexibility and biological characteristics, may be preferable for datasets with fewer records; however, the choice of an appropriate function depends on the dataset characteristics.