计算机技术与发展2025，Vol.35Issue(8)：36-44,9.DOI:10.20165/j.cnki.ISSN1673-629X.2025.0077
概念演化数据流主动学习方法
Active Learning Method for Concept Evolution Data Stream
摘要
Abstract
Research on data stream classification methods focuses on dynamic model update in open environments,aiming to detect and adapt to concept evolution in real-time,continuously changing data streams.Most existing data stream classification methods typically assume that the number of classes in the data stream is fixed and that sample labels can be accessed without restrictions,which is unrealistic in real-world scenarios.An Active Learning Method for Concept Evolution Data Stream(ALM-CEDS)is proposed.The al-gorithm defines a base classifier importance measure based on sample standard deviation and introduces a sample prediction method using weighted prediction probabilities to improve classifier performance.A classifier update strategy based on a hybrid label query approach updates classifiers using hard-to-distinguish samples and those representing the current data distribution.Additionally,a novel class detection method based on the q-nearest neighbor silhouette coefficient of micro-clusters is developed to rapidly identify novel class in the data stream.Comparative experiments on four real-world data streams and five synthetic data streams demonstrate that ALM-CEDS outperforms six existing data stream learning methods in classification performance.关键词
数据流分类/概念演化/主动学习/新类检测/聚类Key words
data stream classification/concept evolution/active learning/new class detection/clustering分类
信息技术与安全科学引用本文复制引用
李艳红,杜江涛,王素格,白鹤翔,李德玉..概念演化数据流主动学习方法[J].计算机技术与发展,2025,35(8):36-44,9.基金项目
国家自然科学基金(62376143,62473241,62476185) （62376143,62473241,62476185）
山西省基础研究计划项目(202203021221001) （202203021221001）
