Performance of Artificial Neural Networks Model under Various Structures and Algorithms to Prediction of Fat Tail Weight in Fat Tailed Breeds and Their Thin Tailed Crosses

Today’s large fat tail lost its importance because of rearing condition and consumers’ demands. Therefore, recording fat tail weight on live animals is important to selecting animals for reduced fat tail weight. The study was conducted to predict the fat tail weight of five different genetic groups of lambs obtained from a mating system between fat-tailed and thin-tailed parents. An Artificial Neural Networks (ANN) procedure was used for prediction performance of different structures (40 levels) and algorithms (5 levels). Eight measurements, including birth type (2 levels), sex (2 levels), breed composition (5 levels), live body weight and four morphological assessments were used as ANN model’s inputs. The results showed that ANN model with adequate structure and algorithm can accurately predict the tail weights and compositions of the studied breeds. Our results indicate that with increase of neurons in first hidden layers, the prediction accu-racies were increase dramatically. Back propagation algorithm (BP) was the best algorithm with higher stable R2 and lower stable root mean squire error (RMSE) in different structures. BP algorithm with 4 and 2 neurons in the first and second hidden layer, respectively, had more ability to predict fat-tail weight in dif-ferent genetic groups. Best ANN model provided 0.962, 0.997 and 0.988 R2 values and 338.156, 43.689 and 117.306 of RMSE for testing, training and the overall data sets, respectively. The study showed that, an ANN model based on the BP algorithm, have high potential to predict fat-tail weight as an important eco-nomic trait in sheep rearing systems.