摘要
针对大多已有基于K近邻和遗传算法的特征选择方法中没有考虑各个特征的重要度不同,并且容易出现过早收敛,特别是局部最优解问题,提出了一种基于自调优自适应遗传算法的WKNN特征选择方法。该方法使用WKNN算法预测样本的类别,为每个特征分配一个权重来衡量特征的分类能力,然后采用自调优自适应遗传算法,对变异率、种群规模和收敛阈值进行参数调整,在迭代进化过程中搜索最优特征权重向量。为了评价该方法的有效性,与已有7种特征选择方法在5个标准数据集上进行了比较。实验结果表明,该方法是有效的,且具有较高的分类性能。
In view of the fact that most of the existing feature selection methods based on K-nearest neighbor and genetic algorithm do not take into account the different importance of each feature and are prone to premature convergence,especially the local optimal solution problem,a weighted K-nearest neighbor feature selection method based on self-tuning adaptive genetic algorithm is proposed in this paper.The method uses the weighted K-nearest neighbor algorithm to predict the category of samples and assigns a weight for each feature to measure the feature’s classification ability,then it uses self-tuning adaptive genetic algorithm to adjust the mutation rate,population size and convergence threshold,and searches for the optimal feature weight vector in the process of iterative evolution.In order to evaluate the effectiveness of this method,it is compared with the existing seven feature selection methods on five standard data sets.Experimental results show that this method is effective and has high classification performances.
作者
陈倩茹
李雅丽
许科全
刘铱龙
王淑琴
CHEN Qianru;LI Yali;XU Kequan;LIU Yilong;WANG Shuqin(College of Computer and Information Engineering,Tianjin Normal University,Tianjin 300387,China)
出处
《计算机工程与应用》
CSCD
北大核心
2021年第20期164-171,共8页
Computer Engineering and Applications
基金
国家自然科学基金(61070089,61703306)
天津市应用基础与前沿技术研究计划项目(15JCYBJC4600,19JCZDJC35100)。
关键词
特征选择
加权K近邻
自调优自适应遗传算法
参数调优
实数编码
feature selection
weighted K-nearest neighbor
self-tuning adaptive genetic algorithm
parameter tuning
real coding
