School of Information and Electronic Engineering, China University of Mining and Technology, Xunzhou, 221116, China.
Sci Rep. 2017 Mar 23;7(1):376. doi: 10.1038/s41598-017-00416-0.
Feature selection is an important data preprocessing technique in multi-label classification. Although a large number of studies have been proposed to tackle feature selection problem, there are a few cases for multi-label data. This paper studies a multi-label feature selection algorithm using an improved multi-objective particle swarm optimization (PSO), with the purpose of searching for a Pareto set of non-dominated solutions (feature subsets). Two new operators are employed to improve the performance of the proposed PSO-based algorithm. One operator is adaptive uniform mutation with action range varying over time, which is used to extend the exploration capability of the swarm; another is a local learning strategy, which is designed to exploit the areas with sparse solutions in the search space. Moreover, the idea of the archive, and the crowding distance are applied to PSO for finding the Pareto set. Finally, experiments verify that the proposed algorithm is a useful approach of feature selection for multi-label classification problem.
特征选择是多标签分类中的一种重要数据预处理技术。尽管已经提出了大量研究来解决特征选择问题，但针对多标签数据的情况却很少。本文研究了一种使用改进的多目标粒子群优化（PSO）的多标签特征选择算法，旨在搜索非支配解集（特征子集）。引入了两种新算子来提高基于 PSO 的算法的性能。一个算子是自适应均匀变异，其动作范围随时间变化，用于扩展群体的探索能力；另一个是局部学习策略，旨在利用搜索空间中稀疏解的区域。此外，归档和拥挤距离的思想被应用于 PSO 以找到 Pareto 集。最后，实验验证了所提出的算法是一种多标签分类问题特征选择的有效方法。
