Comparative Evaluation of ARIMA and ARIMA–ANN Hybrid Models for Forecasting Groundnut Area, Production and Productivity in Surguja District, Chhattisgarh, India
Priyanka Sahu *
Department of Agricultural Statistics and Social Science (L.), Late Dr. RCSD, CARS, Korea, IGKV, Chhattisgarh, India.
Shweta Singh
Department of Agricultural Economics, Late Dr. RCSD, CARS, Korea, IGKV, Chhattisgarh, India.
Sandeep Sonkar
Department of Agricultural Statistics and Social Science (L.), IGKV, Raipur, India.
*Author to whom correspondence should be addressed.
Abstract
This study aims to comparatively evaluate ARIMA and ARIMA–ANN hybrid models to improve forecasting accuracy for planning and policy decisions by examining the time series behavior of groundnut area, production, and productivity in Surguja district, Chhattisgarh, over a 56-year period from 1966–67 to 2021–22. Secondary data were sourced from the EPWRF India Time Series, ICRISAT District-Level Data, the Directorate of Economics and Statistics, and the Directorate of Agriculture, Chhattisgarh. Time series plots and statistical tests, including the Augmented Dickey-Fuller (ADF) test, indicated non-stationarity in the original series, which was addressed through first-order differencing. ARIMA models were then fitted, with ARIMA (0,1,0) selected for area and production, and ARIMA (0,1,1) for productivity, based on minimum AIC and BIC values. Residual analysis and BDS tests confirmed the absence of nonlinear components, suggesting that hybrid ARIMA-ANN models were unnecessary. The ARIMA models alone provided accurate forecasts, as reflected by low RMSE values (0.81 for area, 1.22 for production, and 188.88 for productivity) and MAPE values (10.62 for area, 20.34 for production, and 16.005 for productivity), demonstrating their adequacy for predicting groundnut parameters in Surguja district.
Keywords: ARIMA, ARIMA-ANN, mean absolute percentage error (MAPE), root mean square error (RMSE), forecasting