Optimum Rice Prediction from Conventional, Neural Network and Hybrid models

Abstract

The objective of study was to propose the appropriate model for forecasting the area and production of rice crops in Pakistan. The data for the rice area and production was taken for the agriculture statistics from 1947 to 2020 from the official website, Ministry of Finance, Government of Pakistan. The conventional ARIMA methodology was applied initially to forecast the rice area and production by using the proposed ARIMA(1,1,0) model. Then ARIMA, ETS, TBATS, Artificial Neural Network (ANN) and ARIMA-ETS, ARIMA-TBATS and ARIMA-ANN hybrid model were compared. It was observed that ARIMA(1,1,0) model was best for forecasting the production of rice than all other seven models with the lowest value of RMSE = 401.90 (tons) and MAE = 250.32. For forecasting the rice area, the ARIMA-ANN model was selected because it showed the lowest value of RMSE = 121.89 (Ha) and MAE = 85.82 among all the models. Rice area and production was forecast from 2021 to 2030 by the proposed models, which show that the average prediction of rice production is 7879.96 tons while the rice area would be 3010.82 hectare for the next ten years. It shows that area will be increased by 2.11 percent and rice production will be increased by 11.97 percent upto 2030.. KEYWORDS :Forecasting, ARIMA, TBATS, ETS, ANN.