Learning to Detect the Onset of Disease in Cattle from Feedlot Watering Behavior

Abstract

Intensive farming practices in North America create ideal conditions for disease outbreaks in cattle feedlots. A disease outbreak causes economic loss to the farmer, through the reduced weight gain of sick cattle and the spread of the infection to other cattle. Livestock disease management is a loss-reduction process, where sick cattle are quarantined and treated as early as possible. Disease detection today, however, still relies on behavioral observation and weekly checkups. There is an obvious opportunity for a real-time sensor platform, coupled with intelligent sensor fusion, to significantly improve on the current system. We report on preliminary experiments in developing just such a sensor-fusion system, focusing on a sensor to detect water intake. ROC analysis shows that this sensor does predict disease in cattle. However, a significant skewness in the observed data does degrade the classifier's performance, even when cost-sensitive classification is used.