Comparison of modelling techniques for milk-production forecasting.

Area of Science:

• Dairy Science
• Agricultural Engineering
• Data Science

Background:

• Accurate prediction of daily herd milk yield is crucial for efficient dairy farm management.
• Traditional modeling techniques may have limitations in capturing dynamic production patterns.
• Evaluating advanced modeling approaches is essential for optimizing dairy operations.

Purpose of the Study:

• To assess the suitability of three distinct modeling techniques for predicting total daily herd milk yield.
• To compare the accuracy of nonlinear auto-regressive, artificial neural network, and multiple linear regression models.
• To evaluate model performance across varying forecast horizons, from short-term to a full lactation period.

Main Methods:

• Developed and compared a nonlinear auto-regressive model with exogenous input (NARX), a static artificial neural network (ANN), and a multiple linear regression (MLR) model.
• Utilized three years of historical milk production data from a herd of 140 pasture-based dairy cows.
• Tested model accuracy using forecast horizons of 305 days, and moving piecewise horizons of 50, 30, and 10 days.

Main Results:

• All three models achieved a percentage root mean square error (RMSE) of ≤ 12.03% for the full 305-day lactation prediction.
• The NARX model demonstrated improved accuracy with shorter forecast horizons (RMSE reduced from 8.59% to 5.84% for 305-day to 10-day horizons).
• ANN and MLR models showed less improvement in prediction accuracy as forecast horizons were shortened.

Conclusions:

• The nonlinear auto-regressive model with exogenous input (NARX) is a more accurate alternative for predicting daily herd milk yield compared to ANN and MLR.
• NARX modeling is particularly advantageous for short-term milk yield predictions in pasture-based dairy herds.
• These findings support the adoption of advanced time-series models for enhanced dairy production forecasting.