计算机工程与应用 ›› 2012, Vol. 48 ›› Issue (34): 11-16.

• 博士论坛 • 上一篇    下一篇

一种新的基于约简的多分类器融合算法

刘遵仁1,2,吴耿锋1   

  1. 1.上海大学 计算机工程与科学学院,上海 200072
    2.青岛大学 信息工程学院,山东 青岛 266071
  • 出版日期:2012-12-01 发布日期:2012-11-30

Reduction-based multiple classifiers combination algorithm

LIU Zunren1,2, WU Gengfeng1   

  1. 1.School of Computer Engineering and Science, Shanghai University, Shanghai 200072, China
    2.College of Information Engineering, Qingdao University, Qingdao, Shandong 266071, China
  • Online:2012-12-01 Published:2012-11-30

摘要: 提出了一种新的多分类器融合算法。对特征的提取以约简为基础,按照一定的策略添加若干个属性重要度和特征贡献率大的特征,构成一个融合的特征子集空间;接着借助于kNN的思想,计算测试样本的k个最邻近点的类别百分比,为了提高分类精度,引入了样本相似度测度测试样本与k个最邻近点的相似性,通过设置合适的类别百分比和样本相似度的阈值,最终确定测试样本的类别归属。6个UCI标准数据集的实验分析表明,算法是有效的、可行的。详细分析了不同的约简和不同的阈值对分类精度的影响。

关键词: 邻域粗糙模型, 约简, 属性重要度, 特征贡献率, 融合特征子空间, 样本相似度

Abstract: The feature extraction is based on a reduction, and then to add several features that the value of attribute significance or contribution rate is large according to certain strategy, the feature subset space combined is constituted. With the idea of kNN, to calculate the category percentage of the k-nearest neighbors around the test sample. In order to improve the classification accuracy, the sample similarity measure is introduced to calculate the similarity between the test samples and k-nearest neighbors. By setting the appropriate threshold of the category percentage and the sample similarity, to ultimately determine the category of the test samples. The algorithm’s validity and feasibility have been verified by six multidimensional data sets from UCI. The impact of the different reductions and different thresholds for classification accuracy is analysed detailedly.

Key words: neighborhood rough set model, reduction, attribute significance, feature contribution rate, feature subspace combined, sample similarity