二进制正交电鱼优化算法的特征选择  

Feature selection based on binary orthogonal electric fish optimization algorithm

作　　者：王丹雨 刘昊[1] 丁桂艳[1] WANG Danyu;LIU Hao;DING Guiyan(School of Science,University of Science and Technology Liaoning,Anshan 114051,China)

机构地区：[1]辽宁科技大学理学院,辽宁鞍山114051

出　　处：《辽宁科技大学学报》2022年第3期208-214,共7页Journal of University of Science and Technology Liaoning

基　　金：国家自然科学基金(U1731128);辽宁省教育厅项目(LJKZ0279)。

摘　　要：特征选择是一种机器学习的过程,其通过筛选出信息量大的特征来降低原始数据集的维数,同时使分类特征的精度最大化。为了提高特征选择的能力,本文提出二进制正交电鱼优化(BQOXEFO)特征选择方法。首先,该算法在电鱼优化算法的基础上采用了带有量化的正交交叉策略,提高了算法的收敛速度和精度;其次,在初始化阶段利用转换函数将自变量二进制化,并根据分类的精度和所选特征的数量给出适应度函数,同时,采用K最近邻分类器对数据进行分类。为了验证算法性能,实验基于UCI数据库中的5个基准数据集训练,并与4种算法进行比较。实验和统计分析结果表明,该算法具有良好的性能,它能够很好地完成特征分类任务,使得特征数量最小化,分类性能最大化。Feature selection is a machine learning process that minimizes the dimensionality of the original dataset by filtering out informative features,maximizing the accuracy of categorical features. In order to improve the ability of feature selection,a binary quantization orthogonal crossover electric fish optimization(BQOXEFO)feature selection method was proposed. Firstly,the algorithm adopts an orthogonal crossover strategy with quantization on the basis of the electric fish optimization(EFO),which improves the convergence speed and accuracy of the algorithm. Secondly,during the initialization phase by binarizing the arguments using converting function,the fitness function is proposed according to the classification precision and the number of selected features. At the same time,the K-Nearest Neighbor(KNN)is used to classify the data.To verify algorithm performance,the experiment was trained on 5 benchmark datasets in the UCI database and compared to 4 algorithms. Experimental and statistical analysis results show that the algorithm has good performance,which can well complete the feature classification task so that the number of features is minimized and the classification performance is maximized.

关 键 词：电鱼优化算法 特征选择 智能优化算法 正交设计 

分 类 号：O224[理学—运筹学与控制论]