Input Variable Analysis and Selection for Corn Yield Prediction Using Artificial Neural Network

Abstract

United States of America entered into the advance agricultural practices since 1930. Continuous evolution of agricultural technologies has been witnessed to maximize the production. Efficient production of crop requires effective estimation of yield. Various data-driven models are therefore used to predict the yield. These models rely highly on the retrospective analysis of data. Choosing the right parameters for yield prediction is an essential exercise. This paper evaluates the effective techniques of choosing the right parameters from the data set, finding the correlation between the individual parameters, group of parameters and classifying these parameters which makes an impact on the final yield production. Linear and Nonlinear regression modeling techniques were used for this classification. The selected attributes are then given as an input to an Artificial Neural Network (ANN) to test its prediction capability. Root Mean Squared Error (RMSE), is used as comparison measure. A MatLab program is designed to train and test the models. A dimension reduction approach based on Principal Component Analysis (PCA) gives the best model with minimum RMSE. The crop chosen for our analysis is Corn from the state of Texas. The trained model predicted yield with RMSE of 1206.59 and regression R of 0.63.