Prediction of lifetime milk production using artificial neural network in Sahiwal cattle

Abstract

First lactation records (1493) of Sahiwal cows spread over a period of 40 years (1966–2005) were used to predict lifetime milk production. Data were divided into two sets namely training set (1120 records) and test set (373 records) to compare the accuracy of prediction for lifetime milk production from artificial neural network and multiple regression analysis. The traits considered were age at first calving (AFC), first lactation 305-day or less milk yield (FL305DMY), first lactation length (FLL), first service period (FSP) and first dry period (FDP). The accuracy of prediction of lifetime production from multiple regression analysis was 25.92% from the training set when all the 5 traits were incorporated in the equation, while it was 28.09% for the test data set. An equation incorporating AFC, FL305DMY and FLL was considered to be optimum with an accuracy of prediction of 25.62% for training data set and 27.00% for test data set. The R2– values for prediction the lifetime production by artificial neural networks from training and test set data were 29.81% and 28.88%, respectively; while for optimum equation the corresponding values were 30.04% and 27.78%, respectively. Further, the root mean square errors of prediction were also lower from artificial neural networks in comparison to multiple regression analysis for all prediction equations developed. Higher estimates of accuracy of prediction of lifetime milk yield from artificial neural networks in comparison to multiple regression analysis from overall and optimum equations in both the data sets revealed that this methodology can be used as an alternate approach to predict lifetime milk production in Sahiwal cattle.