GA-ANN HYBRID MODEL FOR PREDICTION OF AREA AND CROP PRODUCTION

The study presents an integrated model that combines Genetic Algorithms (GA) and Artificial Neural Network (ANN)-Multilayer Perceptron (MLP) to optimize the performance of the best fit ANN-MLP model obtained in the study. In artificial neural network, two key issues about the performance are its structure (architecture) and the selection of connection weights that help to minimize the total prediction error. Different neural network weight initialization methods are used before training the network with an aim to avoid slow training of the network and to minimize the error. In the present study, different weight initialization methods for ANN-MLP have been evaluated for the prediction of area and crop production for the crop rice for Barak Valley Zone (BVZ) of Assam, India. The predictive accuracy of the ANN-MLP model developed is optimized with evolving connection weights of ANN-MLP using GA. The predictive accuracy of the developed hybrid model is evaluated with the same dataset considered for ANN-MLP model of the study region. Empirical results obtained demonstrate that the proposed GA-ANN hybrid model can outperform the ANN-MLP for prediction of area and crop production for the crop rice of BV zones of Assam. Further, sensitivity analysis has been carried out with the best GA-ANN hybrid network configuration to identify the most influencing factor in production of rice of Barak Valley zone of Assam.