頁籤選單縮合
題 名 | 分割方法在樹狀迴歸中之比較與研究=Studies on Split Selection Methods for Tree-Structured Regression |
---|---|
作 者 | 史玉山; 吳文揚; | 書刊名 | 中國統計學報 |
卷 期 | 35:2 1997.06[民86.06] |
頁 次 | 頁137-150 |
分類號 | 319.5 |
關鍵詞 | 樹狀迴歸; 不偏; 分割變數; 分割點; Tree-structured regression; Unbiased; Splitting variable; Split point; |
語 文 | 中文(Chinese) |
中文摘要 | 本文主要目的為應用Loh and Shih (1997)所提之假設檢定(hypothesis testing) 方法的原則到樹狀迴歸問題,與 Breiman, et al. (1984) 所提出之徹底搜尋 (exhaustive search) 方法在選取分割變數,分割點和其計算時間上的差異做比較。經由電腦模擬,我們 發現徹底搜尋方法在選取分割變數上會產生偏差,較假設檢定方法要來得不穩健,並且在計 算時間也較費時, 而在選取分割點上,當變數為有序 (ordered) 變數時,假設檢定方法較 徹底搜尋方法精確,而當變數為類別 (categorical ) 變數時,徹底搜尋方法較假設檢定方 法為佳。 |
英文摘要 | In tree-structured regression, a hypothesis testing procedure for split selection is proposed and compared with the exhaustive search method. Based on simulation studies, it is shown that the exhaustive search method can be extremely biased and time consuming with regard to the variables chosen to split the nodes. On the other hand, the new method is almost unbiased. When noise variables are present, the new method is demonstrated to be more robust than the exhaustive search method in choosing splitting variable. When the splitting variable is an ordered variable, the new method is shown to be more accurate in selecting the split point. When the splitting variable is a categorical variable, the exhaustive method is better in choosing the split point. |
本系統中英文摘要資訊取自各篇刊載內容。