Comparison of methods for predicting cow composite somatic cell counts.

Area of Science:

• Veterinary Medicine
• Dairy Science
• Machine Learning

Background:

• Monthly cow composite somatic cell counts (CMSCC) are vital for monitoring udder health and milk quality in dairy herds.
• Current sampling methods are time-consuming and automated tools are costly, creating a need for predictive alternatives.
• Machine learning is increasingly used in mastitis detection, with CMSCC often serving as a predictor or gold standard.

Purpose of the Study:

• To develop a method for predicting cow composite somatic cell counts (CMSCC) between sampling intervals.
• To utilize regularly recorded quarter milk data, such as milk flow and conductivity, for CMSCC prediction.
• To evaluate the efficacy of different machine learning models in predicting CMSCC.

Main Methods:

• Collected quarter-level milk data (flow, conductivity) from 372 Holstein-Friesian cows over 8 weeks.
• Applied machine learning models: generalized additive model (GAM), random forest, and multilayer perceptron (MLP).
• Evaluated models using 7-d and 3-d lagged data, with and without cow number as a predictor, via 5-fold cross-validation.

Main Results:

• Generalized additive model (GAM) emerged as the superior prediction model.
• Multilayer perceptron (MLP) performed comparably to GAM when using less data.
• Including previous CMSCC data significantly reduced prediction error for both GAM and MLP models.

Conclusions:

• Machine learning models, particularly GAM and MLP, show significant potential for predicting CMSCC.
• Regularly recorded quarter milk data can be effectively used to forecast CMSCC, aiding udder health management.
• Previous CMSCC information is a crucial variable for improving prediction accuracy.