Comparison of NNs-ARIMAX and NNs-GSTARIMAX on Rice Price Forecasting in Indonesia
Hasnaq Primageza, Retno Aulia Vinarti, Raras Tyasnurita, Edwin Riksakomara, Ahmad Muklason · 2021
Rice supply contributes the most to the food poverty line in Indonesia since rice is the main food for Indonesian. Therefore, the ups and downs of rice price has great impact to Indonesian. In fact, the national average price of rice in 2020 had fluctuated. This fluctuation was influenced by various factors: the previous period price, the price of grain in the mill, rice stocks, harvested area, rice production, rice consumption, weather, and rice prices in neighboring areas. Therefore, forecasting rice price is carried out in this article. Based on these problems, this study offers a solution to compare methods to predict rice prices in Indonesia. The method is Hybrid NNs-ARIMAX with Hybrid NNs-GSTARIMAX. The selected provinces are West Java, Central Java, East Java, DKI Jakarta, DI Yogyakarta, and Banten. The input variable is historical data on the average price of rice in the period January 1,2008, to December 31,2019 (weekly). The output of this article is the forecast of average rice price and its accuracy performance. The best NNs-ARIMAX model for Banten province is ARIMAX (4,0,5), DKI Jakarta province is ARIMAX (4,1,5), and ARIMAX (1,1,1) for West Java, Central Java, DI Yogyakarta, and ARIMAX (3,0,2) for East Java. The best NNs-GSTARIMAX model is GSTARIMAX (1,1,0)1. This most accurate training-testing is 85:15. There was a 0.17% decrease in MAPE of ANN compared to Hybrid NNs-ARIMAX. Also, 1.09% decrease in MAPE of ANN compared to NNs-GSTARIMAX. This shows that the accuracy of NNs-GSTARIMAX is better than NNs-ARIMAX. So, it can be concluded that the NNs-GSTARIMAX method is better than the NNs-ARIMAX method for predicting rice prices in Indonesia.