摘要
隨著科技的發達，資料產生的數量越來越多，進入了Big Data的時代，透過資料探勘的技術，可以挖掘出更多的知識或是有趣的內容，也能夠做更好的預測。本研究針對醫療資料中的乳癌資料集進行實驗，透過兩個資料大小差異的乳癌資料集探勘後的數據進行分析，根據特徵選取後所取得資料與原始資料做比較，使用單一分類器及多重分類器搭配不同的核心參數進行實驗。透過所得實驗數據，評估那一種分類器及參數的搭配使用，能夠取得較好的效能時間及正確率，如此可使日後研究及預測上能夠有較好的依據，並能夠輔助做出好的決策。 實驗中透過不同核心參數與分類器的搭配得到數據後，以時間及正確率為主要條件排序，找出較好的組合方式，在本研究中，單一分類器使用SVM搭配RBF核心參數，能夠，而多重分類器使用Bagging及Boosting所產生的數據比較後，Boosting的正確性及效能較好。
關鍵字
支援向量機並列摘要
Breast cancer prediction is an important problem in the medical and healthcare communities. In particular, various data mining techniques have been employed to construct the prediction models. Since support vector machines (SVM) are the core machine learning technique and they have shown their outperformance than many other related techniques over many pattern classification problems, very few explore the performances of SVM using different learning functions in breast cancer prediction. Therefore, the aim of this thesis is to use the three well-known kernel functions to develop different SVM classifiers, which are the linear, polynomial, and RBF (radial basis function) kernels, to assess their prediction performances. Moreover, the classifier ensemble techniques based on bagging and boosting are also applied to construct the SVM ensemble classifiers. The experimental results based on two related datasets show that boosting based SVM based on the RBF kernel function performs the best in terms of prediction accuracy and ROC.