结合邻域信息和标记相关性的在线多标记流特征选择算法

现有大多数多标记流特征选择算法在进行特征选择时，往往忽略标记间的相关性，易导致算法预测精度的下降。为解决这一问题，提出一种结合邻域信息和标记相关性的在线多标记流特征选择算法；定义自适应邻域关系解决邻域粗糙集的粒度选择问题，将其推广到多标记学习中；利用互信息计算标记间的相关性得到标记权重；通过邻域粗糙集和标记权重评估特征和标记间的相关性，并设计特征在线重要度分析、在线相关性分析和在线冗余度分析3种指标，以实现在线评价动态候选特征。在7组多标记数据集以及5个评价指标上的实验结果表明，所提算法综合性能较优。

Online multi-label streaming feature selection algorithm via combining neighborhood information and label correlation

Most of the existing multi-label streaming feature selection algorithms tend to ignore the correlation between labels, which easily leads to the decline of prediction accuracy. To address this problem, an online multi-label streaming feature selection algorithm via combining neighborhood information and label correlation is proposed. Firstly, the adaptive neighborhood relationship is defined to solve the problem of granularity selection of neighborhood rough set, and then it is extended to multi-label learning. Secondly, the mutual information is used to calculate the correlation between labels to obtain the label weight. Finally, the neighborhood rough set and label weights are integrated to evaluate the correlation between features and labels. And three metrics, i.e., feature online importance analysis, online relevance analysis and online redundancy analysis, are designed to evaluate the online candidate features. Experiment results on 7 multi-label datasets and 5 evaluation metrics show that the comprehensive performance of the proposed algorithm is effective.