An SVM-like approach for expectile regression

详细信息    查看全文

• 作者：Muhammad Farooq ; ^{muhammad.farooq@mathematik.uni-stuttgart.de} ; Ingo Steinwart ^{ingo.steinwart@mathematik.uni-stuttgart.de}
• 关键词：Asymmetric least square loss ; Expectile regression ; Support vector machines
• 刊名：Computational Statistics & Data Analysis
• 出版年：2017
• 出版时间：May 2017
• 年：2017
• 卷：109
• 期：Complete
• 页码：159-181
• 全文大小：1545 K
• 卷排序：109

文摘

Expectile regression is an interesting tool for investigating conditional distributions beyond the conditional mean. It is well-known that expectiles can be described with the help of the asymmetric least square loss function, and this link makes it possible to estimate expectiles in a non-parametric framework with a support vector machine like approach. For the underlying optimization problem, an efficient sequential-minimal-optimization-based solver is developed and its convergence derived. The behavior of the solver is investigated by conducting various experiments, and the results are compared with the solver for quantile regression and the recent R-package ER-Boost.