摘要
为了提高高维数据流在线分类的准确率,设计一种基于超网络和投影降维的高维数据流在线分类算法。将高维数据流的特征子集建模为超网络模型,算法的学习目标是搜索最优的超边集合,选出判别能力强的特征子集。利用高斯核将高维空间的数据点投影到低维空间,采用梯度下降法计算数据点间的相似性矩阵。基于贝叶斯分类器模型更新机制,动态地学习新到达的数据流,基于学习的结果更新超网络的超边,再利用超网络指导分类器进行分类。仿真结果表明,该算法实现了较高的分类准确率,并且对于噪声也具有较好的鲁棒性。
To improve the classification accuracy of online classification of high dimensional data streams,this paper designs an online classification algorithm for high dimensional data stream based on hypernetworks and projection dimension reduction.It modeled the feature subsets of high dimensional data streams as hypernetworks,the learning objective is to search the optimal hyperedges sets and select the feature subsets with strong discriminant abilities.Gaussian kernel was used to project data points from high dimensional space to low-dimensional space,and gradient descent method was adopted to compute the similarity matrix of data points.Based on the model updating mechanism of Bayes classifier,the arrived data stream was learned dynamically.The hyperedges of hypernetworks was updated based on the learning results,and then the hypernetworks were used to guide the classifier for classification.The simulation experimental results show that this algorithm achieves high classification accuracy and has good robustness to noise.
作者
茹蓓
Ru Bei(School of Computer and Information Engineering,Xinxiang University,Xinxiang 453003,Henan,China)
出处
《计算机应用与软件》
北大核心
2020年第10期278-285,共8页
Computer Applications and Software
基金
河南省软科学研究计划项目(192400410045)。
关键词
超网络
超图
高维数据流
数据流分类
贝叶斯分类器
数据降维
Hypernetworks
Hypergraphs
High dimensional data stream
Data stream classification
Bayes classifier
Data dimension reduction