Using Data Mining Technique to Predict Business Cycle in Taiwan

• Main Authors: Jing-Yi Lyu, 呂憬儀
• Other Authors: Wei-Ping Lee
• Format: Others
• Language: zh-TW
• Published: 2005
• Online Access: http://ndltd.ncl.edu.tw/handle/49559032901572083911

碩士 === 中原大學 === 資訊管理研究所 === 93 === The primary goal of this paper is using data mining technique to construct business predicting model in Taiwan. According to the past reach, the data type of business cycle is asymmetric don’t fit using linear-simple model to analysis. To avoid the problem, we adopt a novel modeling technique, neural network and decision tree, to forecast business cycles. In this paper, we using two data mining approach to forecast business model, and furthermore discussing time serials data processing, different neural network architect and compare decision tree and neural network predicting performance in business cycle to understanding the two data mining techniques’ difference. The empirical results are: 1. In the business forecast problem, this is no different in neural network’s hidden layer but using two hidden layers neural network have better stabling. If you can wait for training time, suggest using two hidden layer neural network to get predicting performance. 2. The time serial data processing as Walczak used has good performance. The extend time serial method has the best performance. 3. Comparing neural network and decision tree, decision tree have better performance than neural network. 4. When we using expansion and contraction to construct business forecasting model can get a very good model. Two data mining techniques also have a good performance and stable model. 5. The finding result, overseas indicators can help business forecasting model’s performance.