摘要
随着数据挖掘技术在工业领域的广泛应用,异常数据检测问题引起更多关注。面对工业数据复杂的数据特点,基于集成学习的异常检测方法展现了明显优势。然而现有集成方法普遍采用静态集成策略,难以充分挖掘基学习器的潜力。在集成异常检测方法中的基学习器融合环节,提出了一种动态选择方法,通过对每个测试样本动态选择最佳基学习器或子集实现最佳融合效果。仿真结果显示,提出的动态选择方法对比现有静态集成方法具有统计意义上的优势。
With the widespread application of data mining technology in the industrial field,the problem of anomaly data detection has attracted more attention.Faced with the complex data characteristics of industrial data,anomaly detection methods based on ensemble learning have demonstrated significant advantages.However,the existing methods generally use static integration,which is difficult to fully exploit the potential of the base learners.With respect to the base learner fusion of outlier ensembles,this paper proposes a dynamic selection method,which dynamically selects the best base learner(s)for each test sample to achieve the best fusion effect.The simulation results show that the proposed method has statistical advantages over the existing static ensembles.
作者
宋艳丽
刘德兵
胡海华
常玉清
SONG Yan-li;LIU De-bing;HU Hai-hua;CHANG Yu-qing(Chengdu Aeronautic Polytechnic,Chengdu Sichuan 610100,China;School of Information Science and Engineering,Northeastern University,Shenyang Liaoning 110819,China)
出处
《计算机仿真》
2024年第12期555-559,共5页
Computer Simulation
基金
国家自然科学基金(62273078)。
关键词
异常数据检测
集成学习
机器学习
Outlier detection
Ensemble learning
Machine learning
