590 0

Robust Estimation of Support Vector Regression with Residual Bootstrap

Title
Robust Estimation of Support Vector Regression with Residual Bootstrap
Author
최원영
Alternative Author(s)
Choi, Won Young
Advisor(s)
차경준
Issue Date
2011-08
Publisher
한양대학교
Degree
Doctor
Abstract
Recently, the request of the optimal design has been increased because the system is so complicated that it requires expensive cost for analysis. In this thesis, more stable approximation model is proposed by applying bootstrap to support vector regression (SVR), which is one of the useful and common approximation modeling systems for prediction. SVR is an efficient system to express the nonlinearity of system relatively well. But, using the SVR does not always yield accurate results, because it is sensitive to the input parameters. In order to overcome this problem, we apply bootstrap technique to SVR. In fact, we apply the bootstrap to the residual of SVR model. The effects of input parameters are relatively eliminated. The performance of the proposed method is evaluated and the valuable results are gained.| 최근의 최적설계는 시스템이 복잡해지고 해석해야 할 때 드는 시간적, 경제적 비용이 증가 되고 있어, 해석 소프트웨어의 이용에 대한 요구가 증가되고 있다. 특히, 근사모델을 이용한 최적설계는 기존의 최적설계 기법의 비효율성을 극복하기 위한 방법으로 많은 연구가 이루어 졌다. 본 논문에서는 최근 많은 각광을 받고 있는 서포트벡터회귀 기법에 잔차 붓스트랩 방법을 적용하여 더욱 안정적인 근사모델을 제안하고자 한다. 서포트벡터회귀기법은 비선형성을 잘 표현해주며 까다로운 계산 없이 근사모델을 만들 수 있다는 장점이 있다. 하지만, 이 기법은 사용자가 입력해야하는 입력변수에 대한 가이드라인이 없고 입력변수에 따라서 그 결과가 상당히 달라진다는 단점이 있다. 이러한 단점을 극복하기 위한 방법으로, 비모수적인 방법으로 대표적이면서 자료에 대한 특별한 가정이 필요하지 않은 붓스트랩을 서포트벡터회귀 기법에 적용하였다. 특히, 붓스트랩 방법중에서도 잔차에 대한 붓스트랩을 적용함으로서 입력변수에 덜 민감한 서포트벡터회귀기법을 제안하였다. 제안된 방법의 효율성을 검증하였으며 유의한 결과를 도출하였다.; Recently, the request of the optimal design has been increased because the system is so complicated that it requires expensive cost for analysis. In this thesis, more stable approximation model is proposed by applying bootstrap to support vector regression (SVR), which is one of the useful and common approximation modeling systems for prediction. SVR is an efficient system to express the nonlinearity of system relatively well. But, using the SVR does not always yield accurate results, because it is sensitive to the input parameters. In order to overcome this problem, we apply bootstrap technique to SVR. In fact, we apply the bootstrap to the residual of SVR model. The effects of input parameters are relatively eliminated. The performance of the proposed method is evaluated and the valuable results are gained.
URI
https://repository.hanyang.ac.kr/handle/20.500.11754/138916http://hanyang.dcollection.net/common/orgView/200000417312
Appears in Collections:
GRADUATE SCHOOL[S](대학원) > MATHEMATICS(수학과) > Theses (Ph.D.)
Files in This Item:
There are no files associated with this item.
Export
RIS (EndNote)
XLS (Excel)
XML


qrcode

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

BROWSE