An intelligent system for livestock disease surveillance

Currently, cattle in feedlots are monitored for disease through a manual, labour-intensive process. Physical checkups are administered approximately once per week, and pen riders are hired to watch over the herd, looking for behaviors that indicate an animal is sick. Contagious diseases thus have considerable freedom to spread before they are first detected, leading to increased morbidity and mortality in the herd. We propose the use of animal-mounted sensors, coupled with an intelligent surveillance system, to automatically and continuously monitor the health of each animal; in essence, this is a case study of designing an intelligent condition-monitoring system, in the form of an inferential sensor. In an empirical trial of the system on an Alberta feedlot, sensor data was used to forecast illness up to seven days in advance. Using an ensemble classifier in the wavelet domain, we obtain a sensitivity of 80.8% and specificity of 80%.