COMPARISON OF ANALYSIS TECHNIQUES FOR ONLINE DETECTION OF CLINICAL MASTITIS

Three techniques were compared for analysis of automatically collected data from the milking parlor. Mammary quarters showing signs of clini cal mastitis were compared with randomly selected healthy quarters. Au tomatic data were analyzed from the milking on which the milkers obser ved clinical mastitis as well as data from the two prior milkings. Ele ctrical conductivity of milk was not corrected for individual cows. Mi lking parlor data were preprocessed so that information on the electri cal conductivity pattern during a milking was retained. Principal comp onent analysis was used to verify whether variation in the data was ca used by mastitis. Performance of logistic regression models for detect ion of clinical mastitis was compared with that of backpropagation neu ral networks. Variation in the quarter data was caused by mastitis. Au tomatic data from infected quarters did not always differ from data fr om healthy quarters, especially from the two prior milkings. The detec tion performance of the logistic regression model was similar to that of the neural networks. When both models were tested on the developmen t data, sensitivity was approximately 75%, and specificity was approxi mately 90% at the milking of mastitis observation. Detection results w ere lower for the prior milkings. Therefore, not all incidences of cli nical mastitis cases could be detected before clinical signs occurred.