Data-Driven Approaches to Multi-Crop Yield Prediction using General Regression Neural Network models

Abstract

Food production is under pressure to fulfill the demands of growing populations. Under this situation, an accurate crop yield prediction helps in agricultural production planning. Due to the inadequate prediction performance by the traditional linear models, the evolution of a crop yield prediction model is necessary. The prediction model should be easy to understand, precise, and requires less time during the training and validation stages. In view of these, the current study emphasizes on developing an adaptive, low-complexity, and precise nonlinear model for prediction of crop yield. An experimental nonlinear relationship between various independent attributes and the crop yield has been derived from the novel General Regression Neural Network (GRNN) model. It is observed that the model shows improvement in the prediction accuracy with mean square error between 0.011-0.017, root mean square error between 0.090-0.109 and R2 value between 0.91-0.94 compared to the corresponding response surface methodology (RSM) model. The GRNN model can be employed to predict different agricultural crops from the same or other geographical regions across the world. By analyzing these results, farmers, agronomists and researchers can gain a better understanding of the interrelationship between different attributes and how they might impact the overall production.