Binary Optimization Using Hybrid Grey Wolf Optimization for Feature Selection
Binary Optimization Using Hybrid Grey Wolf Optimization for Feature Selection
复制标题
DOI：
10.1109/access.2019.2906757
复制
发表时间：
2019-01-01
期刊：
影响因子：
3.9
通讯作者：
Alhussian, Hitham
中科院分区：
文献类型：
--
作者：
Al-Tashi, Qasem;Kadir, Said Jadid Abdul;Alhussian, Hitham
本文提出了一种混合灰狼优化（GWO）和粒子群优化（PSO）的二进制版本来解决特征选择问题。原PSOGWO算法是一种综合了粒子群算法和粒子群算法优点的新型混合优化算法。尽管具有优越的性能，但原始混合方法适用于具有连续搜索空间的问题。然而，特征选择是一个二元问题。因此，提出了一种称为BGWOPSO的混合PSOGWO的二进制版本来寻找最佳特征子集。为了找到最优解，利用基于包装器的欧氏分离矩阵k近邻分类器。为了对所提出的二进制算法进行性能评估，使用了来自UCI存储库的18个标准基准数据集。结果表明，BGWOPSO在精度、选择最优特征和计算时间等性能指标上均显著优于二元GWO （BGWO）、二元PSO、二元遗传算法和模拟退火鲸鱼优化算法。
A binary version of the hybrid grey wolf optimization (GWO) and particle swarm optimization (PSO) is proposed to solve feature selection problems in this paper. The original PSOGWO is a new hybrid optimization algorithm that benefits from the strengths of both GWO and PSO. Despite the superior performance, the original hybrid approach is appropriate for problems with a continuous search space. Feature selection, however, is a binary problem. Therefore, a binary version of hybrid PSOGWO called BGWOPSO is proposed to find the best feature subset. To find the best solutions, the wrapper-based method K-nearest neighbors classifier with Euclidean separation matric is utilized. For performance evaluation of the proposed binary algorithm, 18 standard benchmark datasets from UCI repository are employed. The results show that BGWOPSO significantly outperformed the binary GWO (BGWO), the binary PSO, the binary genetic algorithm, and the whale optimization algorithm with simulated annealing when using several performance measures including accuracy, selecting the best optimal features, and the computational time.