Optimization of fuzzy inference system by using table look-up method to predict white sugar price in the international market

Abstract

The study aims to investigate the most optimum rules of fuzzy inference systems to forecast the white sugar price in the international market. The fuzzy rules are optimized by using a table look-up method. As a comparison, we also employ fuzzy time series methods that developed by Chen, Singh, and Heuristic. The main differences among the four methods are on the inputs determination for prediction and on the algorithm to calculate the prediction. The performance of the method is evaluated using MAPE (Mean Absolute Percentage Error). The MAPE values of white sugar price forecasting yielded by a fuzzy inference system with table lookup are 2.83 % on training data and 7.66 % on checking data. Furthermore, the MAPE values resulted by fuzzy time series model of Chen, Singh, and heuristic are 8.52%, 8.62% and 8.01% on training data, and 6.44%, 6.47%, and 6.44% on checking data, respectively. The table lookup delivers the highest performance on the training data, while it delivers the lowest performance on checking data. However, the fluctuation of its forecasts is more reasonable, since it follows the fluctuation of the actual data, while the other three methods deliver constant forecasts which are not reasonable.