Binary Multi-Objective Grey Wolf Optimizer for Feature Selection in Classification
Binary Multi-Objective Grey Wolf Optimizer for Feature Selection in Classification
复制标题
DOI：
10.1109/access.2020.3000040
复制
发表时间：
2020-01-01
期刊：
影响因子：
3.9
通讯作者：
Alqushaibi, Alawi
中科院分区：
文献类型：
--
作者：
Al-Tashi, Qasem;Abdulkadir, Said Jadid;Alqushaibi, Alawi
特征选择或降维可以看作是一个具有两个目标的多目标最小化问题：最小化特征数目和最小化错误率。尽管特征选择是一个多目标问题，但现有的大多数方法都将特征选择看作一个单目标优化问题。最近，人们提出了多目标灰狼优化算法(Mogwo)来解决多目标优化问题。然而，Mogwo最初是为连续优化问题而设计的，因此不能直接用于解决本质上是离散的多目标特征选择问题。因此，本文提出了一种基于Sigmoid传递函数的二进制MOGO算法--BMOGW-S，用于优化特征选择问题。基于包装器的人工神经网络(ANN)被用来评估所选特征子集的分类性能。为了验证该方法的性能，使用了来自UCI知识库的15个标准基准数据集。将提出的BMOGWO-S算法与MOGWO算法进行了TANH传递函数、非支配排序遗传算法和多目标粒子群算法的比较。结果表明，提出的BMOGWO-S算法能够有效地确定一组非支配解。在大多数情况下，该方法在特征约简和分类错误率方面都优于现有的多目标方法，同时具有较低的计算代价。
Feature selection or dimensionality reduction can be considered as a multi-objective minimization problem with two objectives: minimizing the number of features and minimizing the error rate simultaneously. Despite being a multi-objective problem, most existing approaches treat feature selection as a single-objective optimization problem. Recently, Multi-objective Grey Wolf optimizer (MOGWO) was proposed to solve multi-objective optimization problem. However, MOGWO was originally designed for continuous optimization problems and hence, it cannot be utilized directly to solve multi-objective feature selection problems which are inherently discrete in nature. Therefore, in this research, a binary version of MOGWO based on sigmoid transfer function called BMOGW-S is developed to optimize feature selection problems. A wrapper based Artificial Neural Network (ANN) is used to assess the classification performance of a subset of selected features. To validate the performance of the proposed method, 15 standard benchmark datasets from the UCI repository are employed. The proposed BMOGWO-S was compared with MOGWO with a tanh transfer function and Non-dominated Sorting Genetic Algorithm (NSGA-II) and Multi-objective Particle Swarm Optimization (MOPSO). The results showed that the proposed BMOGWO-S can effectively determine a set of non-dominated solutions. The proposed method outperforms the existing multi-objective approaches in most cases in terms of features reduction as well as classification error rate while benefiting from a lower computational cost.