| 吴倩楠,颜学峰.基于改进最大相关最小冗余的选择性集成分类器[J].高技术通讯(中文),2022,32(1):40~49 |
| 基于改进最大相关最小冗余的选择性集成分类器 |
| Selective ensemble classifier based on improved maximum relevance and minimum redundancy |
| |
| DOI：10.3772/j.issn.1002-0470.2022.01.005 |
| 中文关键词: 选择性集成； 最大相关最小冗余(mRMR)； 特征选择； 正交化； 距离相关系数(DCC) |
| 英文关键词: selective ensemble, maximum relevance and minimum redundancy (mRMR), feature selection method, orthogonalization, distance correlation coefficient (DCC) |
| 基金项目: |
|
| 摘要点击次数: 1479 |
| 全文下载次数: 947 |
| 中文摘要: |
| 在构建选择性集成分类器时，寻找分类准确率高且差异性大的最优分类器子集至关重要。为平衡集成子集中基分类器的准确性和多样性，提出了一种基于改进最大相关最小冗余的选择性集成分类器（ImRMRSEC）。首先，将基分类器对验证集的预测结果视为一个个“特征”，把特征选择的思想扩展到集成分类器的约简问题中，基于最大相关最小冗余准则寻找基分类器子集。其次，引入Gram-Schmidt正交化求取“特征”的等价向量，替代原向量输入最大相关最小冗余算法中，并基于距离相关系数(DCC)衡量相关性。同时，利用序列浮动前向选择方法搜索最优子集。实验结果充分展示了所构建分类器卓越的设计性能。 |
| 英文摘要: |
| When constructing selective ensemble classifiers, it is important to find the optimal subset of classifiers with high accuracy and large differences. In order to balance the accuracy and diversity of base classifiers in the ensemble subset, an improved maximum relevance and minimum redundancy (mRMR)-based selective ensemble classifier (ImRMRSEC) is proposed. Firstly, the prediction of base classifiers on the verification sets are regarded as features, and the idea of feature selection is extended to ensemble pruning. Secondly, the equivalent vector of a feature obtained by Gram-Schmidt orthogonalization is used as the input of mRMR instead of the original vector, and the correlation is measured based on the distance correlation coefficient. Meanwhile, the sequence floating forward selection method is used to search for the optimal subset. The experimental results fully demonstrate the excellent design performance of the constructed classifier. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
