基于特征有效范围的前向特征选择及融合分类算法

Algorithm of Forward Feature Selection and Aggregation of Classifiers Based on Feature Effective Range

特征在不同类别样本之间的重叠区域反映了特征的区分能力.根据特征在各类样本中的有效范围及每一区域样本的分布密度,提出一种基于特征有效范围的前向特征选择及融合分类算法(FFS-ER).该算法采用前向特征搜索策略,在进行特征选择的过程中建立分类模型.为说明该算法的有效性,在8个公共数据集上将其与比较流行的、性能优越的后向特征选择算法SVM-RFE和前向特征选择算法FIM进行比较,实验结果表明该算法所选特征构建的分类模型的分类准确率明显高于FIM算法,且在大多数情况下优于SVM-RFE算法.同时标准偏差的比较说明该算法相对于SVM-RFE和FIM具有较好的稳定性.

Overlapping area of a feature among different groups reflects its discriminative ability. This paper proposes an algorithm （ FFS-ER ） of forward feature selection and aggregation of classifiers based on the effective ranges of features and the distribution den- sity of different group samples. It adopts the forward feature search strategy, and the classification model is established in the process of feature selection. In order to illustrate the effectiveness of the proposed algorithm, it is compared on eight public datasets with SVM- RFE, which is a popular and superior backward feature selection algorithm, and FIM, which is a forward feature selection algorithm. Experimental results show that the feature subsets selected by FFS-ER are more discriminative than those selected by FIM and better than those by SVM-RFE in most cases. Further the comparison on standard deviation implies that FFS-ER is more stable than SVM- RFE and FIM.