摘要： 在线学习方法是用于大规模数据集的、高效且可扩展的机器学习算法。然而，在对多标签数据集进行特征选择时，传统的在线多标签学习方法需要访问数据集的所有特征，当数据集具有较高维度时，这种在线学习方式并不能适用于实际情景。针对多标签数据集的特征选择，在现有研究的基础上，使用二类分解策略，提出基于分解策略的多标签在线特征选择算法。该算法利用稀疏正则化和截取方法进行在线特征选择，降低计算复杂度。实验表明，算法的特征选择性能优于其他多标签在线特征选择算法。

中图分类号： TP305；TP181
文献标识码： A
DOI： 10.19358/j.issn.2097-1788.2022.04.010
引用格式： 张永伟，朱祁，吴永城. 基于分解策略的多标签在线特征选择算法[J].网络安全与数据治理，2022，41(4)：65-71，77.

Online multi-label feature selection algorithm based on binary relevance strategy

Abstract： Online learning method is an efficient and extensible machine learning algorithm for large-scale applications. However, when selecting features for multi-label datasets, the traditional online multi-label learning algorithms need to access all the features of the dataset. When the dataset has a higher dimension, this online learning method cannot be applied to the actual situation. Based on the existing research, a multi-label online feature selection algorithm based on decomposition strategy is proposed by using binary relevance strategy. This algorithm uses sparse regularization and interception methods for online feature selection to reduce computational complexity. The experimental results show that the feature selection performance of the algorithm is better than that of other multi-label online feature selection algorithms.

Key words : feature selection；online learning；multi-label classification；binary relevance strategy