Comparative Efficiency of Lactation Curve Models Using Irish Experimental Dairy Farms Data

Abstract. The objective of this study was to find a well-fitting model to represent the lactation curves of Irish dairy cows over various prediction horizons. Ten representative models from multiple categories including; curve fitting, static regression models and auto-regressive models were used to forecast total daily milk production. The simulations were carried out on a total of six years of daily milk yield records from farms situated in the south of Ireland. The model simulations were set at herd level and a sample herd of 100 cows was randomly selected from the regional herd population. The parity composition of the sample herd was kept identical to that of the regional herd population. The SSE (Sum of Squared Errors) value of the ten lactation models varied substantially, except for the Surface Fitting model. Based on these criteria, the Surface Fitting model had the lowest SSE value (2,097,183 kg and 1,824,517 kg) over the 365-day horizon and the 30-day horizon. The R² value of the Surface Fitting model (0.96 and 0.97) was the highest and the RMSE (Root Mean Squared Error) value of the Surface Fitting model was the lowest (75.8 kg and 70.7 kg), in comparison to the other lactation models. However, for the short-term prediction (10-day horizon), the NARX model proved to be the most accurate forecasting model with the lowest SSE value (1,268,021kg) and the lowest RMSE value (58.9 kg). The results showed the Surface Fitting model and the NARX model provided the most accurate milk production forecast for the sample herd, for long-term (365-day), medium-term (30-day) and short-term (10-day) forecast horizons, respectively.