Constructing Prediction Models for Printed Circuit Board and Integrated Circuit Export Values：Comparison among Regression Analysis, ARIMA, and Soft Computing Approaches

• Main Authors: HUANG, PO-JUI, 黃柏瑞
• Other Authors: Yuehjen E. Shao
• Format: Others
• Language: zh-TW
• Published: 2016
• Online Access: http://ndltd.ncl.edu.tw/handle/6z87b2

碩士 === 輔仁大學 === 統計資訊學系應用統計碩士在職專班 === 104 === Taiwan has a very active electronics manufacturing industry that accounts for a high proportion of the Taiwanese GDP. In recent years, because of difficulty in land allocation, increased labor costs, the “magnet effect” of China, and the westward shift of industries in Taiwan, the GDP has declined and the unemployment rate has increased. Therefore, effective prediction and accurate assessment methods of the production value of the electronics industry that facilitate confirmation of the subsequent export value is a poignant research topic. Because the production values of printed circuit boards (PCBs) and integrated circuits (ICs) in Taiwan are both among the highest in the global market, this study used the export value of these two industries as the prediction targets. This study collected the statistical data of monthly export values of PCBs and ICs published by The Bureau of Foreign Trade, Ministry of Economic Affairs. The predictive power of the following prediction methods were examined and compared: multiple regression, multivariate adaptive regression splines, support vector regression, artificial neural network (ANN), and autoregressive integrated moving average (ARIMA). The results showed that the PCB monthly export value prediction was most accurate when ANN was used, whereas the IC monthly export value prediction was most accurate when ARIMA was used.