Predictions of 305-day milk yield in Iranian Dairy cattle using test-day records by artificial neural network

Abstract

Artificial neural network was used as non-linear models to predict 305-day milk production using test-day records. Test-day records (32475) belong to five recording period, of the first lactation were used in analyses. A total of 75% of records were used for training of back propagation artificial neural network system. The ANN system in this study had 3 layers of input, hidden and output each with 11, 30 and 1 neurons, respectively. The results showed that there was no significant difference between observed and predicted data. R² values ranged from 77% in period 1 to 92% in period 5 when 100 records were used in the analysis. The error coefficients of I²_o, I²_B and I²_E resulted from inadequacy in flexibility and insufficient convergency between direction of changes in the observed and predicted data were reduced as period number was increased. Our results showed that ANN system have ability for reasonable prediction of 305-day milk yield from small number of test–day records in early stages of milk production.