Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.contributor.advisor | 유재근 | * |
| dc.contributor.author | 박유진 | * |
| dc.creator | 박유진 | * |
| dc.date.accessioned | 2024-08-19T16:31:43Z | - |
| dc.date.available | 2024-08-19T16:31:43Z | - |
| dc.date.issued | 2024 | * |
| dc.identifier.other | OAK-000000232151 | * |
| dc.identifier.uri | https://dcollection.ewha.ac.kr/common/orgView/000000232151 | en_US |
| dc.identifier.uri | https://dspace.ewha.ac.kr/handle/2015.oak/268890 | - |
| dc.description.abstract | 본 논문에서는 먼저 새로운 교차 거리 선택 알고리즘을 제안한다. 제안된 선택 알고리즘은 Ye and Weiss (2003)의 부트스트래핑 방법과 유사하게 데이터를 기반으로 하며 성능과 관련하여 강력한 근거를 제시하고 있다. 시뮬레이션 연구에서는 제안된 알고리즘을 통해 선택된 하이브리드 방법이 더 좋은 추정치를 제공하고 있으며 동시에 연산 시간을 크게 감소시킨다는 것을 보여준다. 더 나아가 실제 데이터 분석에서는 제안된 선택 알고리즘이 기존의 부트스트래핑 방법보다 유용한 장점을 가지고 있다는 것을 확인하였다. 또한 자료의 개수 (n) 보다 자료의 변수 (p) 가 더 많은 고차원 데이터에 SDR을 적용하는 새로운 방법을 제안한다. 먼저 공분산 행렬의 비가역 문제를 해결하기 위해 패널티를 부여한 우도 방법을 활용하고 이를 SDR 방법과 결합한다. 제안된 방법은 largen small p 데이터에 대하 SDR 방법의 적용이 가능하도록 하였으며 고차원 자료의 복잡한 패턴을 파악함으로써 통계적 추정의 정확도를 향상시킨다. 추가적으로 시뮬레이션연구와 실제 데이터 분석을 통해 제안된 방법의 우수성을 확인하였다.;Sufficient dimension reduction (SDR) is a promising method as a supervised dimension reduction approach. In this thesis, we first introduce a novel cross-distance selection algorithm. Similar to the bootstrapping method developed by Ye andWeiss (2003), the proposed selection algorithm is data-driven and has a strong rationale for its performance. Numerical studies demonstrate that the hybrid method chosen from our proposed algorithm offers a good estimation quality and reduces the computing time dramatically at the same time. In addition, our real data analysis confirms that the proposed selection algorithm has potential advantages with its practical usefulness over the existing bootstrapping method. Furthermore, we present a new SDR method developed for high-dimensional data. Initially, we employ penalized likelihood methods in order to address non-invertibility of the covariance matrix, combining them with linear sufficient dimension reduction methods. The approach enables the detection of complex patterns within high-dimensional data and enhances the accuracy of statistical estimations. Moreover, we confirm the robustness and effectiveness of the proposed method through simulation studies and real data analysis. Finally, we provide a theoretical perspective on the convergence rate and discuss possible future research challenges. | * |
| dc.description.tableofcontents | I. Introduction 1 A. Research Background & Objectives 1 B. Organization of Thesis 4 II. Sufficient dimension reduction 5 1. SIR (Sliced inverse regression) 8 2. SAVE (Sliced average variance estimation) 10 3. DR (Directional regression) 12 4. pHd (Principal Hessian Directions) 15 5. Covk (Covariance method) 18 III.On Cross-Distance Selection Algorithm for Hybrid Sufficient Dimension Reduction 21 A. Hybrid Sufficient Dimension Reduction 21 1. Candidate methodologies 22 2. Candidate methodologies for hybrid dimension reduction 32 3. Bootstrap method 33 B. Cross-Distance Approach of Hybrid Dimension Reduction 35 1. Narrowing hybrid methods 35 2. Cross-distance selection of α 41 C. Numerical Studies and Real Data Analysis 49 1. Numerical studies 49 2. Real data analysis: Abalone data 51 IV. High dimensional sufficient dimension reduction using sparse matrix inversion 53 A. Sparse-matrix inversion for Sufficient Dimension Reduction 53 1. The graphical Lasso algorithm 54 2. The SCAD algorithm 55 B. The proposed Methods 57 1. The Glasso-DR case 58 2. The SCAD-SIR case 59 3. The SCAD-SAVE case 60 4. The SCAD-DR case 61 C. Numerical Studies and Real Data Analysis 62 1. Numerical studies 62 2. Real Data Analysis 75 D. Theoretical results 79 1. Conditions 79 2. Convergence rate 80 V. Discussion 82 Bibliography 84 ABSTRACT (in Korean) 90 | * |
| dc.format | application/pdf | * |
| dc.format.extent | 3490286 bytes | * |
| dc.language | eng | * |
| dc.publisher | 이화여자대학교 대학원 | * |
| dc.subject | Sufficient dimension reduction | * |
| dc.subject.ddc | 500 | * |
| dc.title | Algorithms for Sufficient Dimension Reduction | * |
| dc.type | Doctoral Thesis | * |
| dc.creator.othername | Park, Yujin | * |
| dc.format.page | v, 90 p. | * |
| dc.identifier.thesisdegree | Doctor | * |
| dc.identifier.major | 대학원 통계학과 | * |
| dc.date.awarded | 2024. 8 | * |