A hybrid forecasting model of cassava price based on artificial neural network with support vector machine technique

Abstract

Thailand is the world's largest exporter of cassava. The cassava prices fluctuate because of many factors such as the production cost, economic condition, and price intervention. Therefore, this research aims to propose a forecasting model of cassava price based on the 11-year data (from 2005 to 2015) obtained from the Thai Tapioca Starch Association and Office of Agricultural Economics. Various techniques were applied for the forecast such as Artificial Neural Network, Support Vector Machine, k-Nearest Neighbor and Hybrid Technique. The statistics used to determine the effectiveness of this model were Mean Absolute Percentage Error (MAPE), Root Mean Squared Error (RMSE) and Mean Squared Error (MSE). The results of this research showed that Hybrid Technique demonstrated the lowest value of error followed by Artificial Neural Network, k-Nearest Neighbor and Support Vector Machine, respectively. Therefore, it could be concluded that using the Hybrid Technique to forecast the price of cassava was better than other techniques and generated the predicted price closest to the actual price.