Improved ARIMAX modal based on ANN and SVM approaches for forecasting rice yield using weather variables

Abstract

An effort has been made to get precise forecast of rice yield through ARIMAX and proposed hybrid models using weather variables. In this article, two hybrid approaches like ARIMAX-ANN and ARIMAX-SVM have been proposed. Firstly, ARIMAX model was fitted for the considered time series data. Rice yield along with weather variables of Aligarh district of Uttar Pradesh have been considered to evaluate the forecasting performance of the proposed hybrid models. The residuals obtained from the fitted model which exhibit nonlinear pattern were fitted employing ANN and SVM. Using the fitted yield values through the hybrid approaches via ANN and SVM, MAPE under ARIMAX (0,1,1)-ANN and ARIMAX (0,1,1)-SVM are estimated to be 0.37 and 1.11, respectively, as compared to 12.18 under ARIMAX (0,1,1) model. Based on the results obtained, we infer that although performance of proposed ARIMAXSVM and ARIMAX-ANN models are close to each other but much superior to the conventional ARIMAX model for the considered data set. Performance of hybrid ARIMAX model is found to be quite encouraging. Yield has also been forecasted up to 2020 on the basis of forecasted rainfall using ARIMAX (0,1,1) model.

Keywords

• ANN
• ARIMAX
• Hybrid ARIMAX
• SVM