Analysis of Staple Food Price Forecasting Results Using Various Approaches

Abstract

Some fields of science often use events in the past as a guide to determine activities that can occur in the future. Forecasting mechanisms are needed to determine uncertain future events. Forecasting results should be able to approach the actual situation with a small error value. The type of data and the forecasting algorithm used can affect the quality of the forecasting results. In this study, we make forecasts to estimate the prices of 11 staple foods in 2017. The data used are time series with a composition of 70% training data and 30% testing data. This study compares Autoregressive Integrated Moving Average (ARIMA), Double Exponential Smoothing (DES), and Support Vector Regression (SVR) with three types of kernels to determine the quality of forecasting results. The test results show that SVR with Radial Basis Function (RBF) kernel provides the best performance compared to other algorithms with an average RMSE of 1520.20 and MAPE of 9.16%. The SVR algorithm with the RBF kernel can predict the price of staple foods close to the actual price in the market.