Mining sensor data to discover clinical mastitis

When cows are milked with an automatic milking system (AMS), clinical mastitis (CM) cannot be detected adequately without using electronic sensing devices. This paper describes approaches to improve automated CM detection in AMS using sensor inputs and data mining. Sensor data and observational CM data, both at quarter level, were collected over two years at nine Dutch AMS farms. Decision-tree induction was used for model development using data from cows that were highly likely to be healthy or that were clearly suffering from CM. The model was validated including quarter milkings with a less clear CM status. A decision-tree was developed with sensitivity of 40% and specificity of 99% using a strict time-window