基于蚁狮算法的元特征选择方法  

Meta-feature selection method based on ant lion optimization algorithm

作　　者：李庚松 刘艺 郑奇斌 秦伟 李红梅 任小广 宋明武 LI Gengsong;LIU Yi;ZHENG Qibin;QIN Wei;LI Hongmei;REN Xiaoguang;SONG Mingwu(Defense Innovation Institute,Beijing 100071,China;Academy of Military Science,Beijing 100091,China;Tianjin(Binhai)Artificial Intelligence Innovation Center,Tianjin 300450,China)

机构地区：[1]国防科技创新研究院,北京100071 [2]军事科学院,北京100091 [3]天津(滨海)人工智能创新中心,天津300457

出　　处：《系统工程与电子技术》2023年第9期2831-2842,共12页Systems Engineering and Electronics

基　　金：科技部科技创新2030-重大项目(2020AAA0104802);国家自然科学基金(91948303);国家自然科学青年基金(61802426)资助课题。

摘　　要：为了提升基于元学习算法选择的性能,提出一种基于蚁狮算法的元特征选择方法。首先,通过鲁棒初始化机制构建初始种群,增强所选元特征子集的鲁棒性。其次,在个体解的搜索过程中应用动态边界策略,增加方法的种群多样性。然后,采用混沌映射变异策略,提升方法的寻优性能,给出方法伪代码并分析时间复杂度。最后,使用130个数据集、150种元特征、8种候选算法和5种性能指标构建分类算法选择问题进行测试实验,分析方法的参数敏感性和机制策略效果,通过准确率、查准率、查全率和F 1分数指标评估并对比方法性能,验证了所提方法的有效性和优越性。To improve the performance of meta-learning algorithm selection,a meta-feature selection method based on ant lion algorithm is proposed.Firstly,an initial population is constructed through a robust initialization mechanism to enhance the robustness of the selected meta-feature subset.Secondly,dynamic boundary strategy is applied in the search process of individual solutions to increase the population diversity of the method.Then,the chaos map mutation strategy is used to improve the optimization performance of the method,give the method pseudocode and analyze the time complexity.Finally,classification algorithm selection problem are constructed using 130 datasets,150 meta features,eight candidate algorithms,and five performance indicators for testing experiments.The parameter sensitivity and mechanism strategy effectiveness of the method are analyzed.The performance of the method is evaluated and compared through accuracy,precision,recall,and F 1 score indicators,verifying the effectiveness and superiority of the proposed method.

关 键 词：元特征选择 蚁狮优化算法 算法选择 元学习 分类 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]