FORECASTING RICE AREA, PRODUCTION AND PRODUCTIVITY IN INDIA USING ARIMA AND HYBRID ARIMA–ANN MODELS

Abstract

Rice is one of the most important food crops in India and plays a crucial role in national food security, agricultural growth and rural livelihoods. Accurate forecasting of rice area, production and productivity is therefore essential for agricultural planning and policy formulation. The present study evaluates the forecasting performance of the Autoregressive Integrated Moving Average (ARIMA) model and a hybrid ARIMA–Artificial Neural Network (ARIMA–ANN) model for India's rice area, production and yield. Annual time-series data covering the period 1961 2023 were analysed. The data from 1961–2018 were used for model development, while observations from 2019 2023 were reserved for out-of-sample validation. Descriptive statistics, normality tests, runs tests, Augmented Dickey–Fuller (ADF) tests and Ljung–Box tests were employed to investigate the statistical characteristics of the series. All three original series were found to be non-stationary, whereas first differencing achieved stationarity. Auto ARIMA identified ARIMA(4,1,0) as the most suitable model for rice area and ARIMA(0,1,1) for both production and yield. Forecast accuracy was assessed using MAE, RMSE, MAPE, R² and MASE. ARIMA performed marginally better than ARIMA–ANN for area and production, whereas the hybrid model showed slightly better performance for yield. Future forecasts for 2024–2033 indicate an overall increase in rice production and productivity, with comparatively slower growth in cultivated area. The findings suggest that productivity improvement will remain an important driver of future rice production growth in India. The study demonstrates the usefulness of both classical and hybrid time-series approaches for agricultural forecasting and provides evidence for informed agricultural planning.