| 拆分特征选择及其在企业信用评估中应用 |
| |
| 引用本文： | 凌健,林成德.拆分特征选择及其在企业信用评估中应用[J].福建工程学院学报,2006,4(4):436-439. |
| |
| 作者姓名： | 凌健 林成德 |
| |
| 作者单位： | 厦门大学自动化系,福建,厦门,361005 |
| |
| 摘 要： | 评估指标体系的选取是企业信用评估的首要问题，它是一个特征选择问题。文章提出了一种针时SVM组合技术的拆分特征选择方法，其主要思想是时SVM组合中的各个分类器分别进行特征选择，再采用不同的特征子集作为各子分类器的输入，进行组合建模与预测。文章从filter和wrapper相结合的思想出发，进行了子分类器的特征选择；之后，针对企业信用评估问题的特点，采用了二叉树结构作为SVM的组合策略。实验表明，拆分特征选择方法能选出规模较小、具有一定差异的关键指标集，提高了模型的分类性能，并且具有计算简单，运行快速的优点。
|
| 关 键 词： | 特征选择 支持向量机(SVM) 企业信用评估 |
| 文章编号： | 1672-4348（2006）04-0436-04 |
| 收稿时间： | 2006-07-04 |
| 修稿时间： | 2006年7月4日 |
Separating feature selection and its application to enterprise credit assessment |
| |
| Ling Jian,Lin Chengde.Separating feature selection and its application to enterprise credit assessment[J].Journal of Fujian University of Technology,2006,4(4):436-439. |
| |
| Authors: | Ling Jian Lin Chengde |
| |
| Abstract: | The selection of the evaluating index system is the key to enterprise credit assessment,which is essentially a feature selection problem.A separating feature selection approach concerning the combination of SVMs(support vector machine) is proposed,whose basic idea is to execute feature selection on each SVM in the combination and use the different selected feature subsets as the inputs.The composite feature selection based on filter and wrapper is used in the selection process of each SVM.The SVMs are then combined using binary tree structure adapted to the characteristic of enterprise credit assessment.Experiment shows that separating feature selection can select feature subsets with small scale and diversity and improve the classification ability of the model and reduce the computing time and complexity. |
| |
| Keywords: | feature selection support vector machine(SVM) enterprise credit assessment |
| 本文献已被 CNKI 维普 万方数据 等数据库收录！ |
