Hybrid statistical model for forecasting production of maize crop in Karnataka State, India

Abstract

The present study was carried out to apply the Hybrid statistical model specifically Auto Regressive Integrated Moving Average (ARIMA) combined with Time-Delay Neural Network (TDNN) on Maize production in Karnataka state for the period of 1962 to 2022. The ARIMA (3, 1, 2) model were found to be best forecasted model for Maize production by criteria, which having highest R2, low Akaiki Information criteria (AIC) and low Root Mean Square Error (RMSE). Consequently, a nonlinear TDNN model was applied to fit the residuals. For Maize production forecasting, the TDNN (4-15-1) model with 4 input delays, 15 hidden neurons, and 1 output neuron demonstrated low RMSE and low Mean Absolute Percentage Error (MAPE) ensuring high accuracy. The ARIMA (3,1,2) model with TDNN (4-15-1) model on ARIMA residuals was found to be the best forecasting model for Maize production of Karnataka state, India. Finally, the hybrid model ARIMA with TDNN was improved the forecasting accuracy. The forecasted results for Maize production for Karnataka state were shown an increasing trend.