Application of a neural network to analyse on-line milking parlour data for the detection of clinical mastitis in dairy cows

As part of research towards an on-line mastitis detection system, standard back-propagation neural networks to classify healthy and clinical mastitic quarters were explored. The usage of back-propagation neural networks is discussed. A neural network for clinical mastitis detection is presented. This network used automatically collected quarter electrical conductivity data as input. The network was trained with 17 healthy and 13 clinical mastitic quarters. All healthy and 12 of 13 mastitic quarters were classified correctly after training. The trained neural network predicted 34 of 38 healthy quarters correctly in different evaluation data sets. For mastitic quarters, related data had to be used, and 21 of 38 mastitic quarters were classified correctly. We concluded that a back-propagation neural network could indeed separate healthy from clinical quarters in an experimental setting. Further development should include the use of different input parameters and comparisons with other analysis techniques.