面向大规模特征选择的自监督数据驱动粒子群优化算法  

A self-supervised data-driven particle swarm optimization approach for large-scale feature selection

作　　者：黎建宇 詹志辉[1] LI Jianyu;ZHAN Zhihui(School of Computer Science and Engineering,South China University of Technology,Guangzhou 510006,China)

机构地区：[1]华南理工大学计算机科学与工程学院,广东广州510006

出　　处：《智能系统学报》2023年第1期194-206,共13页CAAI Transactions on Intelligent Systems

基　　金：国家重点研发计划项目(2019YFB2102102);国家自然科学基金面上项目资助(62176094)。

摘　　要：大规模特征选择问题的求解通常面临两大挑战:一是真实标签不足,难以引导算法进行特征选择;二是搜索空间规模大,难以搜索到满意的高质量解。为此,提出了新型的面向大规模特征选择的自监督数据驱动粒子群优化算法。第一,提出了自监督数据驱动特征选择的新型算法框架,可不依赖于真实标签进行特征选择。第二,提出了基于离散区域编码的搜索策略,帮助算法在大规模搜索空间中找到更优解。第三,基于上述的框架和方法,提出了自监督数据驱动粒子群优化算法,实现对问题的求解。在大规模特征数据集上的实验结果显示,提出的算法与主流有监督算法表现相当,并比前沿无监督算法具有更高的特征选择效率。Large-scale feature selection problems usually face two challenges:1)Real labels are insufficient for guiding the algorithm to select features,and 2)a large-scale search space encumbers the search for a satisfactory high-quality solution.To this end,in this paper,a novel self-supervised data-driven particle swarm optimization algorithm is proposed for large-scale feature selection,including three contributions.First,a novel algorithmic framework named selfsupervised data-driven feature selection is proposed,which can perform the feature selection without real labels.Second,a discrete region encoding-based search strategy is proposed,which helps the algorithm to find better solutions in a large-scale search space.Third,based on the above framework and method,a self-supervised data-driven particle swarm optimization algorithm is proposed to solve the large-scale feature selection problem.Experimental results on datasets with large-scale features show that the proposed algorithm performs comparably to the mainstream supervised algorithms and has higher feature selection efficiency than state-of-the-art unsupervised algorithms.

关 键 词：特征选择 大规模优化 粒子群优化算法 进化计算 群体智能 数据驱动 自监督学习 离散区域编码 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]