Time-series models on somatic cell score improve detection of mastitis

Autor: Karen Helle Sloth, E. Norberg, P. L⊘vendahl, I. R. Korsgaard
Rok vydání: 2008
Předmět:
Zdroj: Norberg, E, Korsgaard, I R, Sloth, K H M N & Løvendahl, P 2008, ' Time-series models on somatic cell score improve detection of matistis : ', Acta Agriculturae Scandinavica, Section A-Animal Science, vol. 58, no. 4, pp. 165-169 . https://doi.org/10.1080/09064700802621143
ISSN: 1651-1972
0906-4702
DOI: 10.1080/09064700802621143
Popis: In-line detection of mastitis using frequent milk sampling was studied in 241 cows in a Danish research herd. Somatic cell scores obtained at a daily basis were analyzed using a mixture of four time-series models. Probabilities were assigned to each model for the observations to belong to a normal "steady-state" development, change in "level", change of "slope" or "outlier". Mastitis was indicated from the sum of probabilities for the "level" and "slope" models. Time-series models were based on the Kalman filter. Reference data was obtained from veterinary assessment of health status combined with bacteriological findings. At a sensitivity of 90% the corresponding specificity was 68%, which increased to 83% using a one-step back smoothing. It is concluded that mixture models based on Kalman filters are efficient in handling in-line sensor data for detection of mastitis and may be useful for similar applications to decision support systems Udgivelsesdato: December In-line detection of mastitis using frequent milk sampling was studied in 241 cows in a Danish research herd. Somatic cell scores obtained at a daily basis were analyzed using a mixture of four time-series models. Probabilities were assigned to each model for the observations to belong to a normal "steady-state" development, change in "level", change of "slope" or "outlier". Mastitis was indicated from the sum of probabilities for the "level" and "slope" models. Time-series models were based on the Kalman filter. Reference data was obtained from veterinary assessment of health status combined with bacteriological findings. At a sensitivity of 90% the corresponding specificity was 68%, which increased to 83% using a one-step back smoothing. It is concluded that mixture models based on Kalman filters are efficient in handling in-line sensor data for detection of mastitis and may be useful for similar applications to decision support systems
Databáze: OpenAIRE
Externí odkaz:
Nepřihlášeným uživatelům se plný text nezobrazuje