Files
Abstract
A particular type of “Artificial neural network (ANN)”, viz. Multilayered feedforward artificial neural
network (MLFANN) has been described. To train such a network, two types of learning algorithms,
namely Gradient descent algorithm (GDA) and Conjugate gradient descent algorithm (CGDA), have
been discussed. The methodology has been illustrated by considering maize crop yield data as response
variable and total human labour, farm power, fertilizer consumption, and pesticide consumption as
predictors. The data have been taken from a recently concluded National Agricultural Technology
Project of Division of Agricultural Economics, I.A.R.I., New Delhi. To train the neural network,
relevant computer programs have been written in MATLAB software package using Neural network
toolbox. It has been found that a three-layered MLFANN with (11,16) units in the two hidden layers
performs best in terms of having minimum mean square errors (MSE) for training, validation, and test
sets. Superiority of this MLFANN over multiple linear regression (MLR) analysis has also been
demonstrated for the maize data considered in the study. It is hoped that, in future, research workers
would start applying not only MLFANN but also some of the other more advanced ANN models, like
‘Radial basis function neural network’, and ‘Generalized regression neural network’ in their studies.