Effects of Learning Parameters and Data Presentation on the Performance of Backpropagation Networks in Milk Yield Prediction

In recent years much attention has been given to using backpropagation neural networks to solve real-world problems in the field of agriculture. In building a back propagation network, various architectures, learning rates, learning rules, momentum and the method of presentation of the input data may be used. Since, in employing neural networks, the user is not expected to understand thoroughly their internal functioning, it is a common practice to use the default configurations, including learning parameter values, provided by commercially available softwares. However, in order to optimize learning, convergence speed, and predictive ability, it is, at times, necessary to adjust some or all of these parameters. While variations in performance due to modifying the net architectures were frequent, few studies have considered the effect of other factors on the learning of neural nets.[...