Use of test-day records to predict first lactation 305-day milk yield using artificial neural network in Kenyan Holstein–Friesian dairy cows

Abstract

The study is focused on the capability of artificial neural networks (ANNs) to predict next month and first lactation 305-day milk yields (FLMY305) of Kenyan Holstein–Friesian (KHF) dairy cows based on a few available test days (TD) records in early lactation. The developed model was compared with multiple linear regressions (MLR). A total of 39,034 first parity TD records of KHF dairy cows collected over 102 herds were analyzed. Different ANNs were modeled and the best performing number of hidden layers and neurons and training algorithms retained. The best ANN model had one hidden layer of logistic transfer function for all models, but hidden nodes varied from 2 to 7. The R ² value for ANNs for training, validation, and test data were consistently high showing that the models captured the features accurately. The R ², r, and root mean square were consistently superior for ANN than MLR but significantly different (p > 0.05). The prediction equation with four variables, i.e., first, second, third, and fourth TD milk yield, gave adequate accuracy (79.0%) in estimating the FLMY305 from TD yield. It emerges from this study that the ANN model can be an alternative for prediction of FLMY305 and monthly TD in KHF.