基于节点评估与最大类间方差的孤立森林异常值检测  

Isolation Forest Outlier Detection Based on Node Evaluation and Otsu

作　　者：严爱军 和世潇[1,2] 汤健 YAN Aijun;HE Shixiao;TANG Jian(Faculty of Information Technology,Beijing University of Technology,Beijing 100124,China;Engineering Research Center of Digital Community,Ministry of Education,Beijing 100124,China;Beijing Laboratory for Urban Mass Transit,Beijing 100124,China)

机构地区：[1]北京工业大学信息学部,北京100124 [2]数字社区教育部工程研究中心,北京100124 [3]城市轨道交通北京实验室,北京100124

出　　处：《北京工业大学学报》2024年第10期1188-1197,共10页Journal of Beijing University of Technology

基　　金：国家自然科学基金资助项目(62073006);北京市自然科学基金资助项目(4212032)。

摘　　要：针对孤立森林(isolation forest, iForest)无法有效检测局部异常值且异常值分数阈值难以精确设定的问题,提出一种基于节点评估(node evaluation, NE)与最大类间方差(Otsu)的iForest异常值检测方法。首先,在样本评估过程中将节点深度与相对质量同时引入评分机制,使算法对全局和局部异常值敏感;然后,为了准确设定分数阈值,采用Otsu自适应设定异常值分数阈值;最后,在不同数据集上验证所提方法的有效性。实验结果表明,该方法可以有效兼顾全局和局部异常值的检测,提高iForest检测异常值的准确性。Isolation forest(iForest)cannot effectively detect local outliers and the outlier score threshold is difficult to be precise,therefore,an isolation forest outlier detection method based on node evaluation(NE)and maximum between-class variance(Otsu)was proposed.First,the scoring mechanism was introduced into the node depth and relative mass at the same time during the sample assessment process,so that the algorithm was sensitive to global and local outliers.Afterwards,to accurately set the score threshold,the Otsu method was used to adaptively determine the outlier score threshold.Finally,the effectiveness of the proposed method was verified on different datasets.Results show that the proposed method can effectively balance the detection of global and local outliers,and can improve the accuracy of detection of outliers in isolation forests.

关 键 词：孤立森林(isolation forest iForest) 异常值检测 最大类间方差(Otsu) 节点评估(node evaluation NE) 分数阈值 节点深度 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]