2007-08-012024-05-16https://scholars.lib.ntu.edu.tw/handle/123456789/668029摘要：依據各試驗單位的特性建立樣本分類法則廣泛應用在各研究領域, 隨著現代生物科技(如微晶片) 的快速發展, 更顯示對大量數據建立分類法則的重要性。本計畫提出建構大量數據之樣本分類法則的方法如下: 利用本質相關係數篩選數個可明確判別樣本群集的變數, 再藉由貝氏累加迴歸樹對所選變數建立最佳分類法則。本研究目的除了探討利用上述方法建立樣本分類法則的可行性外, 將利用資料模擬的方式比較其與傳統分類法之優劣。最後, 此一方法將實際應用在實際癌症基因表現資料的分類。本研究之長期目標期望藉由分類法則的建立, 針對病人情況提供臨床醫療的建議, 達到個人化醫療的終極目標。<br> Abstract: The problem of classification is to assign objects to one of the mutually exclusive subgroups in the population based on the object's characteristics. With the development of microarray technology, the inquiry of classification methods for massive amount of data has risen. In this project we propose an approach that combines the coefficient of intrinsic dependence (CID) and the model of Bayesian additive regression trees (BART). The CID first select essential features. The BART follows to build an accurate classifier for the objects. We will compare the CID-BART approach with conventional methods by simulations. The analyses of actual gene expression data from carcinoma studies will be included as well. The ultimate goal of this research is the system of personalized therapy based on the classification results.本質相關係數貝氏累加迴歸樹特徵篩選分類法則微陣列Coefficient of Intrinsic DependenceBayesian Additive Regression TreesFeature SelectionClassificationMicroarray