Using machine learning methods to predict dry matter intake from milk mid-infrared spectroscopy data on Swedish dairy...

Area of Science:

• Animal Science
• Agricultural Engineering
• Spectroscopy

Background:

• Accurate dry matter intake (DMI) is crucial for dairy production but difficult to measure in group-fed herds.
• Traditional DMI estimation relies on animal parameters like age, weight, and lactation stage.
• Milk mid-infrared spectra (MIRS) offer a potential alternative for estimating DMI.

Purpose of the Study:

• To compare three modeling approaches for predicting DMI using MIRS data.
• To evaluate the accuracy of DMI prediction models based on different data combinations.
• To identify the optimal method for DMI prediction in dairy cows.

Main Methods:

• Collected DMI and MIRS data from a research herd over five years.
• Utilized partial least-squares regression, support vector machine regression, and random forest regression.
• Compared prediction accuracy (R²) using MIRS alone versus MIRS combined with milk yield and lactation stage data.

Main Results:

• Milk MIRS alone provided low to moderate DMI prediction accuracy (R² = 0.07-0.40).
• Incorporating additional variables significantly improved model accuracy and reduced prediction error.
• Partial least-squares regression demonstrated superior performance for MIRS-based DMI prediction.
• Combining MIRS, milk yield, and concentrate DMI data achieved good prediction accuracy (R² = 0.52-0.65).

Conclusions:

• DMI prediction from MIRS is enhanced by including milk yield and lactation stage data.
• Partial least-squares regression is the most effective method for developing DMI prediction models from MIRS.
• Accurate DMI estimation in dairy cows can be achieved using a combination of spectral and performance data.