Modeling cow somatic cell count using sensor data as input to generalized additive models.

Area of Science:

• Animal Science
• Dairy Science
• Veterinary Medicine

Background:

• Somatic cell count (SCC) is a key indicator of udder health in dairy cows.
• Predicting composite milk SCC (CMSCC) between routine samplings can enhance udder health surveillance.
• Automatic milking rotary systems generate extensive sensor data relevant to milk quality.

Purpose of the Study:

• To investigate the potential of using sensor data from automatic milking rotaries to model cow CMSCC.
• To determine the optimal data window (number of days prior) for predicting CMSCC.
• To identify key sensor variables for accurate CMSCC modeling.

Main Methods:

• Data from 372 Holstein-Friesian cows in a German dairy herd were analyzed.
• Sensor data including quarter conductivity, milk flow, and milking parameters were collected.
• Generalized Additive Models (GAM) were employed for variable selection and multivariable modeling, incorporating lagged data up to seven days.

Main Results:

• A model incorporating sensor data from the seven days preceding sampling, including the sampling day, best explained CMSCC.
• Using data from only three days prior to sampling was found to be sufficient for modeling CMSCC.
• Variables related to quarter conductivity and combined quarter conductivity data were significant predictors of CMSCC.

Conclusions:

• Cow CMSCC can be accurately modeled using routinely collected sensor data from milking robots.
• Predictive modeling of CMSCC using automated milking system data offers a valuable tool for dairy herd management and udder health monitoring.
• The findings suggest a practical approach to improve individual cow udder health surveillance through sensor data analysis.