半监督学习算法下数字化信息归并分类仿真  

Simulation of Digital Information Merging and Classification under Emi-Supervised Learning Algorithm

作　　者：胡建平[1] 严永康 HU Jian-ping;YAN Yong-kang(Nantong University,Jiangsu Nantong 226019,China)

机构地区：[1]南通大学,江苏南通226019

出　　处：《计算机仿真》2023年第12期502-505,562,共5页Computer Simulation

基　　金：江苏省高校哲学社会科学研究项目(2019SJA1473)。

摘　　要：由于数字化信息具有无限性、高维度以及不平衡等特征,难以准确分类,为此提出基于半监督学习的数字化信息归并分类算法。检测原始数字化信息中的概念漂移数据,滤除噪声数据,降低对后续分类结果精度的影响;利用半监督学习训练K均值聚类算法,利用训练后算法训练数据块,构建基础分类器,通过对目标函数求解获得最优聚类中心;构建基于SDClass算法的归并分类器,计算每个数据块类标签的估计值,以及估计值与簇中心间距离,找出最近的簇,将对应的数据块划分到该簇中,实现数字化信息的归并分类。选取6种不同类型的数据集对所提方法展开实验测试,结果表明,所提方法针对不同类型的数据集均可实现高精准分类,且具有较高的分类效率。At present,digital information has the characteristics of infinity,high dimension and imbalance,it is difficult to classify it accurately.Therefore,a merging sort algorithm for digital information based on semi-supervised learning was proposed.Firstly,we detected the concept drift data in original information,and filtered out the noise data,thus reducing the impact on the accuracy of subsequent classification results.Secondly,we used semi-super⁃vised learning to train the K-means clustering algorithm and then trained the data blocks.Meanwhile,we built a bas⁃ic classifier.Moreover,we calculated the objective function,and thus obtained the optimal clustering center.Further⁃more,we built a merge classifier based on the SDClass algorithm to calculate the estimated value of the class label in each data block,as well as the distance between the estimated value and cluster center,thus obtaining the nearest cluster.Finally,we partitioned the corresponding data blocks into this cluster,thus achieving the merge sort of digital information.Six different types of datasets were selected for testing the proposed method.Simulation results show that the proposed method can achieve high-precision classification for different types of datasets,with high classification efficiency.

关 键 词：半监督学习 数字化信息 归并分类 均值聚类算法 簇中心 

分 类 号：TP384[自动化与计算机技术—计算机系统结构]