改進支援向量資料描述中特徵分類準確度之研究

A Study of Improving Feature Classification Results for Support Vector Data Description

王啟凱(Chii-Kae Wang)

指導教授 ： ARRAY(0xc8dfbf0)

中原大學/工學院/機械工程研究所/博士(2009年)

https://doi.org/10.6840/CYCU.2009.00900

摘要

多數的樣本識別研究大多著重於分類與推論領域，而事實上資料描述分類的重要性不亞於前述兩項工作。資料描述的本質不在於如何有效分割重疊或混雜在一起的資料，而是如何明確地判斷出哪些資料是屬於同一個群體(Group)或類別(Class)，意即資料描述或識別的意義在於離群值(Outlier)或邊界值(Boundary)的判別，離群值或邊界值愈少愈明確，有助提升影像判讀的準確度。支持向量資料描述(Support Vector Data Description, SVDD)脫胎於支援向量機(Support Vector Machine , SVM)，能提供有效之資料描述識別結果。惟常囿於樣本資料有限，影響識別結果之準確度。
本研究提出利用極大-極小區間法(Max-min range method)與平均數標準差法(Mean-standard method)作為人工離群值取樣之計算基礎，人工產生圍繞於目標資料(Target set)的離群樣本，並利用 摺 次交叉驗證( -fold times cross validation)方法，我們可以得到懲罰因子參數 與間距寬度參數 的最佳( , )組合結果。最後，我們以UCI標準測試資料庫對兩種方法進行測試，以驗證其有效性。

關鍵字

支援向量 ； 特徵分類

並列摘要

Most pattern recognition tasks deal with classification and regression problems. But the data domain description problem is as important as them. In domain description, the task is not to part with overlapped or mixed objects, but to judge them into the same group or class. It means if we can find the boundary around the target data closely, we can get better accuracy of image judgment. Support Vector Data Description (SVDD) is inspired by the Support Vector Machine (SVM), and can provide an effect accuracy of data domain description. But the accuracy is blundered by the amount of samples.
In this dissertation, we utilize max-min range method and mean-standard method to generate outlier objects around the target data artificially. By -fold times cross validation method, we can get the best ( , ) combination. At last, we use the UCI Machine Learning Dataset Repository to validate the effect of two methods.

並列關鍵字

Support Vector ； Feature Classification

參考文獻

[1] N. J. Pizzi, R. A. Vivanco, and R. L. Somorjai, “EvIdent: A functional magnetic resonance image analysis system,” Artificial Intelligent Med., vol. 21, no. 1-3, pp. 263–269, Jan. 2001.

[3] G. Fumera, F. Roli, and G. Giacinto, “Reject option with multiple thresholds,” Pattern Recognition, vol. 33, no. 12, pp. 2099–2101, Dec. 2000.

[4] S. Roberts and L. Tarassenko, “A probabilistic resource allocating network for novelty detection,” Neural Computing, vol. 6, no. 2, pp. 270–284, Mar. 1994.

[5] D. Y. Yeung and C. Chow, “Parzen-window network intrusion detectors,” in Proceeding 16th ICPR, Quebec, QC, Canada, vol. 4, pp. 385–388, Aug. 2002.

[6] B. Scholkopf, J. C. Platt, J. Shawe-Taylor, A. J. Smola, and R. C. Williamson, “Estimating the support of a high-dimensional distribution,” Neural Computing, vol. 13, no. 7, pp. 1443–1471, Jul. 2001.