摘要
针对高维数据引起的“维数灾难”问题,设计了一种基于神经网络树和人工蜂群优化的高维数据聚类算法.首先,设计了改进的二元人工蜂群优化算法,以封装式方法最大化径向基函数网络的准确率,以过滤式方法最小化特征的冗余度;然后,基于每个特征子集的样本集训练径向基函数网络,构建以径向基函数网络为节点的神经树;最终,采用门网络将连接的类簇分离,获得最终的聚类结果.基于高维数据集和低维数据集均完成了仿真实验,结果表明本算法对于高维数据集实现了较高的聚类准确率.
Focusing on the“curse of dimensionality”problem caused by high dimensional data,a neuron network tree and artificial bee colony optimization based clustering algorithm for high dimensional data is designed.Firstly,an improved binary artificial bee optimization algorithm is designed,the accuracy of radial basis function network is maximized by a wrapper method,the feature redundancy is minimized by a filter method;then,a radial basis function network is trained by samples corresponding to each feature,a neuron tree that each node consists of a radial basis function network is constructed;finally,the gating network is adopted to separate the jointed clusters to output the final results.Simulation experiments are done based on both high dimensional datasets and low dimensional datasets,the results show that the proposed algorithm realizes good clustering accuracy to high dimensional datasets.
作者
吉珊珊
Ji Shanshan(Department of Computer Enginneering,Dongguan Polytechnic,Dongguan 523808,China)
出处
《南京师大学报（自然科学版）》
CAS
CSCD
北大核心
2021年第1期119-127,共9页
Journal of Nanjing Normal University(Natural Science Edition)
基金
东莞市科技局项目(2020507156694)
东莞职业技术学院横向课题(202021189).
关键词
高维数据
神经网络树
人工蜂群优化
聚类算法
特征选择
high dimensional data
neuron network tree
artificial bee optimization
clustering algorithm
feature selection
