AN INVESTIGATION INTO THE USE OF MACHINE LEARNING FOR DETERMINING ESTRUS IN COWS

A preliminary investigation of the application of two well-known machi ne learning (ML) schemes - C4.5 and FOIL - to detection of oestrus in dairy cows has been made. This is a problem of practical economic sign ificance as each missed opportunity for artificial insemination result s in 21 days lost milk production. Classifications were made on normal ised deviations of milk volume production and milking order time serie s data. The best learning scheme was C4.5, which detected 69% of oestr us events, albeit with an unacceptably high rate of ''false positives' ' (74%). Further work based on the use of a progesterone assay to prov ide a more accurate oestrus reference is suggested, along with the inc lusion of more monitored variables and an analysis of their relative c ontributions to the learning process.