Building Cabbage Price Forecasting Models by Data Mining Methods -- An Application of Taiwan Open Data

• Main Authors: SHIE,YU-MIN, 謝馭旻
• Other Authors: LI, JUNG-BIN
• Format: Others
• Language: zh-TW
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/u3cgbq

碩士 === 輔仁大學 === 統計資訊學系應用統計碩士班 === 105 === This study applies cabbage transaction information from Taiwan open data to build price forecasting data mining models. Being one of the vegetables with the top trading volume, cabbage is regarded as the price indicator of overall vegetable trading. A precise and instantaneous price forecasting model may help the authority and farmers for their decision making. This study adopts three data mining models, support vector regression, random forest, and neural network, to build independent price forecasting models with different combination of model parameters. Based on the model outcomes, the random forest model has the best RMSE for Pingtung area. For all the remaining areas, the support vector regression model has the best performance. Additionally, this study applies moving window to split testing and training data sets of the random forest model which outperforms other two models. The result confirms that with the adoption of moving window, the performance of the random forest model can be further improved.Finally, between moving window and the traditional method of comparison, the results of the counties and cities in the RMSE gap is not large, but there are many gaps in the MAPE, so the moving window performance is better than the original cutting.