| 研究生: |
洪劭宇 Hung, Shao-Yu |
|---|---|
| 論文名稱: |
迴歸條樣的節點選擇方式 A knot selection scheme for regression splines |
| 指導教授: |
黃子銘
Huang,Tzee-Ming |
| 口試委員: |
翁久幸
Weng,Chiu-Hsing 鄭宇翔 Cheng,Yu-Hsiang |
| 學位類別: |
碩士
Master |
| 系所名稱: |
商學院 - 統計學系 Department of Statistics |
| 論文出版年: | 2026 |
| 畢業學年度: | 115 |
| 語文別: | 中文 |
| 論文頁數: | 34 |
| 中文關鍵詞: | 無母數迴歸 、迴歸樣條 、空間自適應迴歸樣條 、節點選擇 、史坦無偏風險估計 |
| 外文關鍵詞: | Nonparametric Regression, Regression Splines, Spatially Adaptive Regression Splines, Knot Selection, Stein’s Unbiased Risk Estimate |
| 相關次數: | 點閱:47 下載:0 |
| 分享至: |
| 查詢本校圖書館目錄 查詢臺灣博碩士論文知識加值系統 勘誤回報 |
Regression spline 在無母數迴歸中被廣泛應用,為了改善固定節點方法難以有效捕捉局部尖峰或跳躍結構問題,Zhou and Shen (2001) 提出 Spatially Adaptive Regression Splines(SARS),透過局部樣條近似與節點新增、移動、刪除步驟提升節點配置效果,並以 Stein's Unbiased Risk Estimate(SURE)作為模型選擇準則進行節點更新。
本研究以 SARS 為基礎,改良其節點搜尋步驟內容,透過初始節點篩選與樣本大小相關之參數調整,強化其有限樣本下節點搜尋,模擬以 Bumps 函數為主要測試函數,並輔以 Blocks、HeaviSine 與 Doppler 函數進行比較,想了解不同的節點篩選對估計表現的影響。
Regression splines have been widely applied in nonparametric regression due to their flexibility and local approximation capability. To address the issue that fixed-knot methods may fail to capture local spikes or discontinuities, Zhou and Shen (2001) proposed Spatially Adaptive Regression Splines (SARS), which has better performance in knot configurations through local spline approximation and iterative knot addition, relocation, and deletion procedures. Stein's Unbiased Risk Estimate (SURE) is adopted as the model selection criterion for updating knots.
This study is based on the SARS framework and aims to improve its knot searching procedure. By incorporating an initial knot screening strategy and sample-size-dependent parameter adjustments, the proposed approach enhances knot identification under finite-sample settings. Simulation studies are conducted using the Bumps function as the primary test function, together with the Blocks, HeaviSine, and Doppler functions for comparison, in order to investigate the effects of different knot screening strategies on estimation performance.
摘要 i
Abstract ii
Contents iii
圖目錄 vi
表目錄 vii
1 緒論 1
2 文獻回顧 3
2.1 無母數迴歸模型 3
2.2 Regression Splines 與 B-spline 4
2.3 固定節點迴歸條樣 5
2.4 Penalized Splines 與 Smoothing Splines 6
2.5 Adaptive Knot Selection 6
2.6 Spatially Adaptive Regression Splines (SARS) 7
3 研究方法 8
3.1 資料模型假設 8
3.2 本研究與原始SARS方法之差異 8
3.3 模型選擇準則: Stein's Unbiased Risk Estimate(SURE) 9
3.4 演算法參數設定 10
3.5 初始節點選取方法 12
3.5.1 局部模型與檢定統計量 13
3.5.2 多尺度搜尋與候選節點集合 13
3.5.3 間距篩選節點 14
3.6 節點搜尋流程 15
3.6.1 節點新增 15
3.6.2 節點移動 17
3.6.3 節點刪除與停止準則 17
3.6.4 停止準則 18
3.7 模型效能評估指標 18
4 實驗模擬 19
4.1 模擬設計 19
4.2 模擬結果 20
4.2.1 Bumps函數下的ISE比較 20
4.2.2 節點數量統計結果 21
4.2.3 演算法SURE值收斂情形 22
4.2.4 最佳、中位數與最差模擬結果比較 23
4.3 動態調整參數ISE比較 27
4.3.1 SURE 懲罰係數C(n)設定比較 27
4.3.2 初始節點子樣本數設定比較 27
4.4 其他測試函數補充分析 28
4.4.1 ISE 比較結果 28
4.4.2 ISE 統計摘要 28
4.4.3 模型複雜度與收斂情形 29
4.4.4 中位數模擬結果之 fitted curve 比較 29
4.4.5 補充分析小結 30
4.5 結果討論 30
5 結論與建議 32
5.1 結論 32
5.2 建議 32
References 34
Donoho, D. L. and Johnstone, I. M. (1994). Ideal spatial adaptation by wavelet shrinkage. Biometrika, 81(3):425–455.
Eilers, P. H. C. and Marx, B. D. (1996). Flexible smoothing with b-splines and penalties. Statistical Science, 11(2):89–121.
Friedman, J. H. (1991). Multivariate adaptive regression splines. The Annals of Statistics, 19(1):1–67.
Stein, C. (1981). Estimation of the mean of a multivariate normal distribution. The Annals of Statistics, 9(6):1135–1151.
Zhou, S. and Shen, X. (2001). Spatially adaptive regression splines and accurate knot selection schemes. Journal of the American Statistical Association, 96(453):247–259.
全文公開日期 2031/07/28