NOVEL WRAPPER APPROACH FOR VARIABLE SELECTION IN SUPPORT VECTOR MACHINE
NOVEL WRAPPER APPROACH FOR VARIABLE SELECTION IN SUPPORT VECTOR MACHINE
复制标题
DOI：
10.5109/2203031
复制
发表时间：
2016-12
期刊：
影响因子：
--
通讯作者：
Satoru Koda;R. Nishii;Yuta Fukasawa
中科院分区：
文献类型：
--
作者：
Satoru Koda;R. Nishii;Yuta Fukasawa
支持向量机（SVM）是一种高效的分类机器学习方法。在本文中，我们提出了两个使用包装方法的支持向量机变量选择准则。这些标准衡量每个变量对目标函数的贡献。变量重要性是在测量量的基础上量化的。该方法对所有变量的重要性进行评估，无需进行递归计算，具有较高的计算效率。将其应用于多个人工数据集和真实数据集，结果优于现有方法。
Support vector machine (SVM) is an e ffi cient machine learning method for classification. In this paper, we propose two variable selection criteria for SVM that use wrapper methods. The criteria measure the contribution of each variable for a target function. The variable importance is quantified on the basis of the measured amount. The methods have high computational e ffi ciency because they evaluate the importance of all variables without recursive calculations. They were applied to several artificial and real-world data sets, and their results were superior to those of existing methods.