|Title||Detecting Clinical Mastitis using Sensor Data from Automatic Milking Systems|
|Supervisor||Ad Feelders, Henk Hogeveen|
|ECTS||7.5 or 15|
|Related Course(s)||Advanced Data Mining|
|Description||Although clinical mastitis (an inflammation of the udder) has a low prevalence, it is one of the most costly diseases in the dairy industry.
When milking automatically (with a milk robot), sensors measure all kinds of milk characteristics (e.g., electrical conductivity) during the milking process.
This information is used by clinical mastitis detection models to produce mastitis alert lists.
These lists report cows that are suspected to have CM and that need a visual check by the dairy farmer.
If clinical mastitis is confirmed, an antibiotic treatment is often started.
Current CM detection models show a sensitivity that is too low, meaning that too many true cases are not detected.
Also specificity, referring to the number of non-cases that will be classified as healthy,
needs improvement as farmers experience the large amount of false alerts on the mastitis lists as annoying.
So far, we lack information what the best algorithm is to analyze clinical mastitis field data,
which are imbalanced, noisy, and incomplete by nature, and in addition comes up with a model that fulfills some predetermined
requirements of a clinical mastitis detection model (e.g., a short time window in which a model should alert for clinical mastitis).
The goal of this project is to analyze sensor data collected at ten Dutch Dairy Farms to detect CM using different algorithms to
find out the best method to handle this field data. Used algorithms should be data-driven (e.g., decision tree induction, neural network),
and should fulfill a number of predetermined requirements.
|Special Note||This project is in collaboration with the Department of Farm Animal Health, Utrecht University|