The use of artificial neural networks to diagnose mastitis in dairy cattle

Abstract

The use of milk sample categorization for diagnosing mastitis using Kohonen's self-organizing feature map (SOFM) is reported. Milk trait data of 14 weeks of milking from commercial dairy cows in New Zealand was used to train and test a SOFM network. The SOFM network was useful in discriminating data patterns into four separate mastitis categories. Several other artificial neural networks were tested to predict the missing data from the recorded milk traits. A multi-layer perceptron (MLP) network proved to be most accurate (R² = 0.84, r = 0.92) when compared to other MLP (R² = 0.83, r = 0.92), Elman (R² = 0.80, r = 0.92), Jordan (R² = 0.81, r = 0.92) or linear regression (R² = 0.72, r = 0.85) methods. It is concluded that the SOFM can be used as a decision tool for the dairy farmer to reduce the incidence of mastitis in the dairy herd