A Comparative Study of Activation Functions and Training Algorithm of NAR Neural Network for Crop Prediction

Abstract

The proposed study in this paper provides long-term crop prediction for Tamilnadu, India. Nonlinear Autoregressive (NAR) Neural Network (NN) with different parameter settings has been used to facilitate the correct quality and quantity of crop production. At the core of this study is to compare the effect of training algorithms (such as trainlm, trainbr, trainscg, traincgf, trainbfg, traincgf) and activation functions (such as tansig, elliotsig, logsig and purelin) in the performance of the crop yield forecasting model. This study showed that activation functions elliotsig and tansig with the training algorithm trainbr of NARNN delivered the most promising results based on the smallest error between actual and predicted value compared to the other activation and training functions of NARNN.