-
- Nakamura Masatoshi
- Dainippon Sumitomo Pharma Co., Ltd.
-
- Shimokawa Toshio
- Yamanasi University
-
- sakamoto Wataru
- Osaka University
-
- Goto Masashi
- Biostatistical Research Association
Bibliographic Information
- Other Title
-
- Lasso調整型確率化平衡樹木による回帰解析
- Lasso チョウセイガタ カクリツカ ヘイコウ ジュモク ニ ヨル カイキ カイセキ
Search this article
Abstract
Random Forest (RF) is one of tree-structured approaches, such as classification and regression trees, involved in ensemble learning method to predict outcome more precisely. In this paper we proposed an adjusted RF based on lasso (lasso-RF) for further predictive performance in regression analysis. In particular, we integrated lasso which use one of shrinkage estimators into the tree-structured model of RF. Practically we carried out two case studies and a small scale simulation with factors which influence prediction of outcome such as sample size {100, 200, 400}, methods {RF, lasso-RF} and bootstrap re-sampling times {100, 200, 400}, and evaluated predictive performance. Our case studies suggested that lasso-RF decreased the mean squared errors between true and estimate outcomes less than original RF. By our simulation study, we evaluated influencing factors on the mean squared errors based on analysis of variance with above three factors and their interaction, and showed that the factor of methods as fixed effect was significant at level 0.05 with proportion of variation 29.2%.
Journal
-
- Bulletin of the Computational Statistics of Japan
-
Bulletin of the Computational Statistics of Japan 26 (1), 17-31, 2013
Japanese Society of Computational Statistics
- Tweet
Details 詳細情報について
-
- CRID
- 1390001204380229760
-
- NII Article ID
- 110009686647
-
- NII Book ID
- AN10195854
-
- ISSN
- 21899789
- 09148930
-
- NDL BIB ID
- 025179590
-
- Text Lang
- ja
-
- Data Source
-
- JaLC
- NDL
- CiNii Articles
- KAKEN
-
- Abstract License Flag
- Disallowed