基于模糊粒度条件熵与改进萤火虫算法的特征选择方法  

Feature Selection Method Based on Fuzzy Granularity Conditional Entropy and Improved Firefly Algorithm

作　　者：于浩淼 杨志勇 江峰[1] YU Haomiao;YANG Zhiyong;JIANG Feng(College of Information Science&Technology,Qingdao University of Science and Technology,Qingdao 266061,China)

机构地区：[1]青岛科技大学信息科学与技术学院,山东青岛266061

出　　处：《沈阳大学学报（自然科学版）》2024年第5期401-410,共10页Journal of Shenyang University：Natural Science

基　　金：国家自然科学基金资助项目(62172249);山东省自然科学基金资助项目(ZR2022MF326)。

摘　　要：提出了一种基于模糊粒度条件熵与改进萤火虫算法(firefly algorithm,FA)的特征选择方法FS_FGCEIFA。将模糊粗糙集中的知识粒度与λ-条件熵结合,提出模糊粒度条件熵这一新的信息熵模型;将模糊粒度条件熵应用于FA中,提出一种基于模糊粒度条件熵的适应度函数;采用引力搜索算法中粒子惯性质量和万有引力的计算策略来调整FA中萤火虫的亮度和吸引力,并且将基于引力搜索的自适应步长因子融入FA的位置更新中。在多个UCI数据集以及软件缺陷预测数据集上的实验表明,FS_FGCEIFA能够获得更好的分类性能。A feature selection method,FS_FGCEIFA,was proposed based on fuzzy granularity conditional entropy and improved firefly algorithm(FA).By combining the knowledge granularity and λ-conditional entropy in fuzzy rough sets,a new information entropy model-fuzzy granularity conditional entropy was put forward;the fuzzy granularity conditional entropy was applied to FA,and a fitness function based on fuzzy granularity conditional entropy was proposed;the calculation strategy of particle inertia mass and universal gravity in the gravity search algorithm was adopted to adjust the brightness and attractiveness of fireflies in FA,and the adaptive step size factor based on gravity search was integrated into the position update of FA.Experiments on multiple UCI datasets and software defect prediction datasets show that FS_FGCEIFA could achieve better classification performance.

关 键 词：特征选择 萤火虫算法 引力搜索算法 模糊粒度条件熵 模糊粗糙集 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]