6.2. Early detection of metabolic disorders in dairy cows by using sensor data

The transition period is a crucial period for the dairy cow. A negative energy balance results in an increased risk of metabolic disorders such as milk fever, ketosis and left displaced abomasum. Detection of these metabolic diseases may be improved by using sensors. The Dutch Smart Dairy Farming project has examined the potential of sensor systems. A detection model has been developed and tested. Sensors were installed on a practical dairy farm (300 cows, automatic milking system) for automated measurement of milk yield, milk composition (fat, protein), visits to the milking robot (rewarded and unrewarded), concentrate intake, visits to the concentrate feeder (rewarded and unrewarded), activity, rumination activity and body weight. It is known from the literature that most of these variables are influenced by metabolic disorders. Sensor measurements were aggregated to the daily level and used in the detection model to generate three types of alert: (1) level alert: the value was outside a confidence interval (based on a moving average and standard deviation for preceding values), or (2) trend alert: the change in successive values was different from what might be expected, or (3) index alert: given on specific days, such as the day of calving, in several situations, e.g. when the weight loss was much higher than average weight loss. An alert for metabolic disorder was generated when the number of alerts exceeded a predefined threshold. These metabolic alerts were compared with the reference data to estimate the model performance. The results from the detection model (sensitivity and specificity) depended on the model settings. The results confirmed the potential of a model-based approach for the detection of cows suffering from a metabolic disorder. However, it was difficult to reach an appropriate specificity level (99% or higher). Selecting a smart combination of variables could improve the results.