多角度标签结构和特征融合的多标签特征选择

Multi-viewpoint label structure and feature fusion for multi-label feature selection

特征选择是提高多标签分类性能的一种关键技术,一些特征选择算法在选择与标签相关的特征时没有从多个角度考虑标签结构,致使好的特征未能被选取,影响分类性能.为此,提出一种多角度标签结构和特征融合的多标签特征选择MLSFF.该算法的主要特点是:1)从三个角度考虑标签结构,提取三个重要的特征子集. 2)融合提取的三个特征子集将整个特征空间划分为三个不同重要性的子空间. 3)根据各子空间的重要性,分别设置每个子空间选取比率,并从各子空间中选取冗余度较低的特征,合并成所要的特征子集.多个数据集的实验结果表明,该算法与对比算法相比具有明显的分类效果.

Feature selection is a crucial technique to improve the performance of multi-label classification.Some feature selection algorithms fail to consider the label structure from multi-viewpoint when selecting features related to labels,which results in the failure of selecting good features and affects the classification performance.In this paper,we propose a multi-viewpoint label structure and feature fusion for multi-label feature selection(MLSFF).Its main features are:1)It extracts three important feature subsets by considering label structure from three viewpoints.2)It fuses the three feature subsets extracted to divide the whole feature space into three subspaces of different importance.3)According to the importance of each subspace,it sets the selection ratio of each subspace respectively,and selects features with low redundancy from each subspace to merge them into the desired feature subset.The experimental results of multiple data sets show that this algorithm has obvious classification effection compared with the comparison algorithm.