Early Detection Method for Subclinical Mastitis in Auto Milking Systems Using Machine Learning

Abstract

Bovine mastitis is an inflammation of the udder or mammary gland and dairy farmers must control its occurrence to prevent economic losses. The introduction of auto milking systems makes management of farms and udder health more efficient and auto detection systems for common diseases in dairy farms, which are implemented auto milking systems and detect the diseases based on some measurements while milking, are needed. In this study, we propose a novel model for subclinical mastitis detection. Our dataset was collected from dairy farms in Japan and labeled using risk values calculated by a commercially available milk analyzer based on lactate dehydrogenase (LDH) in order to train our model. Several measurements that can be obtained from any auto milking system, such as electrical conductivity in milk, were used as time series features. The models were trained using machine learning (a support vector machine or random forest) and their performances were compared. Our model detects the onset of subclinical mastitis with an accuracy of 81% in terms of sensitivity and 46% precision. In addition, some cases of subclinical mastitis can be detected earlier than when using an alert system based on LDH. Our model can be expected to be improved and utilized in real dairy farms.