Prediction of First Lactation Milk Yield on the Basis of Test Day Yield Using Artificial Neural Network versus Multiple Linear Regression in Gir Cows

Abstract

The test- day model is a method of choice for the study of milk yield traits and this method is very important in countries like India where herd size is generally smaller and lacking well-established milk recording system. The present study was aimed to predict first lactation milk yield on the basis of test day yield in Gir cows and comparison was made between the relative efficiency of Artificial Neural Network (ANN) and Multiple Linear Regression (MLR) models. First lactation records of 513 Gir cows sired by 75 bulls spread over a period of 34 years (1981 to 2014), maintained at Cattle Breeding Farm, Junagadh were used for the study. The data of monthly test-day milk yield (MTMY) was divided into seven sets. ANN was used with back propagation Bayesian regularization (BR) algorithm and MLR was used with backward elimination method. The accuracy of prediction of first lactation milk yield in MLR was lower than the accuracy of ANN in all the test data sets. The Root Mean Square Errors (RMSE) of prediction were lower in ANN as compared to MLR. The optimum equation had total four variables (test days) viz. TD2 to TD5 for prediction of First Lactation 305-Days Milk Yield (FL305DMY). This equation gave an accuracy of prediction of 76.02% by MLR and 87.69% by ANN model as early as 125^th days of lactation i.e. 5^th monthly test day.