Estimation of production and yield of pulses using ARIMA-ARNN model

Several time series generated from agriculture can be effectively modelled using various time-series modelling techniques such as ARIMA (Box-Jenkins) modelling technique, State-Space modelling technique, Structural Time Series modelling and other time series modelling depend on the properties of the given time series. Modelling and related forecasting for thetime serieswere performed using Autoregressive Moving Average (ARIMA), Autoregressive Neural Network (ARNN) and ARIMA-ARNN hybrid models. First, to maintain the stationarity property of the data (1950–51 to 2017–18)as a necessary step, the datasetwas tested, and thefirst order difference series were considered for modelling using the Box-Jenkins approach. ARIMA (0, 1, 1) were found suitable for the production and yield databased on the least value of Schwarz-Bayesian Criterion (SBC). Secondly, Autoregressive Neural Network (ARNN) of orderARNN (2, 2) wasselected for both the dataset. Lastly, ARIMA (0, 1, 1)-ARNN (4, 6) for both production and yield were found suitable. All the three models were tested for their forecast accuracy using Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE). Accordingly, the ARIMA-ARNN hybrid model was found to be best as compared to the individual ARIMA and ARNN model. Based on the ARIMA-ARNN model, the forecasting of the production and yield for the year 2050 was found to be 35.84 million tonnes and 1062.01 kg/ha, respectively of pulses in India.