A Study on Optimal Tree using Advanced Bumping Procedure

• 발행기관: 한국자료분석학회
• 저자명: 이성건(Seong Keon Lee)
• 간행물 정보: 『Journal of The Korean Data Analysis Society (JKDAS)』Vol.8 No.5, 1645~1655쪽, 전체 11쪽
• 주제분류: 자연과학 > 통계학
• 파일형태: PDF
• 발행일자: 2006.10.30

영문 초록

Breiman(1996) showed how one can use the bootstrap for the more primary purpose of producing a better estimator in decision tree. Unfortunately, the averaging process that produces the bagged estimate θ destroys any simple structure that is present in the original estimate θ . Furthermore, Tibshirani(1997) proposed Bumping procedure using bootstrap samples to provide candidate models for the model search. This has the advantage that it preserves the structure of the estimator while still inducing stability. However, the misclassification error of Bumping tree is generally greater than the ensemble tree’s. Therefore, we propose advanced Bumping procedures that are interpretable and more efficient, that is two-stage Bumping and weighted Bumping.

목차

1. Introduction

2. Bumping Procedure

3. Advanced Bumping Procedure

4. Numerical study

5. Conclusion and Discussion

References

키워드

bootstrap decision tree two-stage Bumping weighted Bumping