以資料為基礎的迴歸模型建構分析－以房屋價格為例

On Some Regression Model Construction and Analysis Based on Housing Data

李仁棻(Ren-Fen Lee) ； 毛麗琴(Li-Zin Mao) ； 黃登源(Deng-Yuan Huang)

《智慧科技與應用統計學報》 7卷1期 (2009/06) Pp. 23-40

https://doi.org/10.29807/JTITAS.200906.0002

摘要

本文主要以房屋價格資料為基礎，結合決策樹(Decision tree)建構線性迴歸模型之案例分析，並研究模型之診斷與變數選擇程序問題。將成對變數利用CART (Classification and Regression Tree)法的工具，逐次搜索各種屬性水準組合，進行因子水準的合併，而降低模型中因子之水準數目，適切掌握因子間的相關與互動，再進一步作資料分析並建構模型後再精簡變數，使線性迴歸模型能更有效反應母體的結構。文中以資料為基礎，找出控制各種相關與互動的變數、並降低屬性之水準數；提出多重決策程序(Multiple decision procedures)並兼顧暫定模型假設的檢驗，及進一步診斷模型選擇的適切性；最後再提出選擇變數之準則。

關鍵字

CART ； 影響點 ； 多重決策程序 ； 假設檢定 ； 變數選擇

並列摘要

In regeression models, the selection procedures are usually selecting suitable models. It would be foolish to choose a model simply because some model selection procedure presents it to you as a good model because the biased data. Considering such as model interpretability and the consistency of the data with model assumptions may dictate choosing some suitable model. For this purpose we first to reduce the levels of factors under controlling the interactions. We will propose some multiple decision procedures to select a good model based on housing data. The results will be efficiently inference the population structure.

並列關鍵字

CART ； influential data ； multiple decision procedures ； testing hypotheses ； variable selection