不同機器學習分類演算法之比較-以統計製程數據為例

The Comparison with Various Machine Learning classification Algorithms - Case Study of Statistical Process Data

張嘉惠(Jia-Huey Chang)

指導教授 ： 鄭春生 教授

元智大學/工程學院/工業工程與管理學系/碩士(2006年)

摘要

資料探勘技術中的分類功能主要是利用歷史經驗將資料歸類，並加以研究各類別的特徵，以利日後作為預測未分類資料特性之依據。此外，分類法為目前最常被企業用來作為商業決策與預測的資料探勘工具之一。由文獻中得知，資料的特性將會影響分類演算法的判斷結果與改變其敏感度；且目前尚未有一種演算法能適用於所有的資料型態。因此，若在進行分類研究前，能對不同演算法所適用的資料特性加以瞭解，應能顯著提升分類演算法的辨識績效。

本研究之目的在於探討資料型態對不同分類演算法辨識績效的影響，並以製程數據作為本研究的分析資料，實際依分類的結果進行特性分析。本研究將比較四種分類技術，分別為：決策樹C5及CART演算法與類神經網路中的徑向基網路法 (Radial Basis Function network)及倒傳遞網路法(Back-propagation Network)。

關鍵字

分類；決策樹；類神經網路

並列摘要

In data mining technology, the purpose of classification is to forecast new evidence with historical classified characteristics. This classification method is usually adopted by enterprises to make a strategic decision nowadays.

According to the literature, the data characteristic affects the judgment result of classification algorithm. Its sensitivity is influenced by data types, so no algorithm suits the all types. Therefore, it’s helpful to the identification performance if we can judge which algorithm is used by data characteristic. The purpose of this paper is to explore which data characteristic will determine the identification performance for the various classification algorithms. We evidence with process data and analyze the features by the classification result. In this paper, we compare the four classification algorithms, decision tree (C5 and CART) and artificial neural network (Radial Basis Function Network and Back-Propagation Network).

並列關鍵字

Classification ； Decision Tree ； Artificial Neural Network,

參考文獻

3. Deboeck, G.., and Kohonen, T., Visual Exploration in Finance with Self-Organizing Maps. Springer-Verlag, London (1998).

5. Frawley, W., Piatetsky, G., and Matheus, C., Knowledge discovery in databases: an overview, AI Magazine, Fall, 213-228 (1998).

6. Han, J. and Kamber, M., Data Mining: Concepts and Techniques, Morgan Kaufmann, CA, USA (2001).

10. Johnson, R. A., and Wichern, D. W., Applied Multivariate Statistical Analysis, Prentice Hall (2002).

11. Kiang. M. Y., “A comparative assessment of classification methods” Decision Support Systems:, 35, 441-454 (2003).