Inducing inference rules for the classification of bovine mastitis

Mastitis, an inflammation within the mammary gland, can be a challenge to diagnose, based on dairy production records and on-farm observations, by the novice in the absence of culture results. The objective of this research was to establish a set of inference rules to classify bacterial causes of mastitis in dairy herds. Inductive inference utilizes specific observations about objects and an initial inductive hypothesis to obtain an inductive assertion that accounts for the observations. Interactive Dichotomizer 3 technology was used to create decision rules for classifying mastitis. It develops decision trees for discrete classification of data. The attributes adopted in this paper were lactation number, days in milk, persistency of milk production, number of severe cases, and average somatic cell count during the current lactation. The mastitis classification categories were divided into four logical groups based on common mastitis control schemes: no growth, contagious, environmental, and others. Causative organisms were determined by bacteriologic culturing of milk samples collected from 30 herds studied in the US. A total of 1253 sample records were selected to generate and test the rules. The established rules were tested using a randomly selected groups of 50 and 250 cows. The accuracy of mastitis classification using induced rules as compared with actual culture results ranged from 58 to 61%. The induced rules showed good potential for presumptive classification of mastitis based on production records. These diagnostic rules are easily implemented, lend themselves to computer analysis, and are easily understood by field practitioners and producers.