基于自适应邻居图的离群点检测方法  

Outlier detection method using adaptive neighbor graph

作　　者：缑鹏飞 宋承云 Gou Pengfei;Song Chengyun(College of Computer Science&Engineering,Chongqing University of Technology,Chongqing 400054,China)

机构地区：[1]重庆理工大学计算机科学与工程学院,重庆400054

出　　处：《计算机应用研究》2023年第11期3309-3314,共6页Application Research of Computers

基　　金：重庆理工大学研究生教育高质量发展行动计划资助项目(gzlcx20232062)。

摘　　要：离群点检测的目标是识别数据集中与其他样本明显不同的个体,以便检测数据中的异常或异常状态。现有的方法难以有效应对复杂、非线性分布的数据,并且面临参数敏感性和数据分布多样性的问题。为此,现提出一种新型图结构——自适应邻居图,以边为导向,通过迭代的方式对数据进行特征提取,并计算近邻可达度对离群点进行识别,减小了参数的影响,同时可适用于不同分布类型的数据。为了充分验证其性能,将该方法在多个合成与真实数据集上同其他方法进行了比较分析。实验结果表明,该方法在所有19个数据集中平均排名第一,在保持高精度的同时表现出稳定性。The objective of outlier detection is to recognize individuals within a dataset who exhibit marked dissimilarity from other samples,commonly referred to as outliers,to detect anomalies or aberrant states in the data.Numerous existing outlier detection methods struggle to handle complex,nonlinearly distributed data,and suffer from the problems of parameter sensiti-vity and diverse data distribution.To address these challenges,the proposed method introduced a novel graph structure called the adaptive neighbor graph.The adaptive neighbor graph was edge-oriented and performed feature extraction iteratively.Also,this method calculated the neighbor reachability to identify outliers.This approach mitigated the influence of parameter and could handle data with various distributions.To evaluate its performance,the proposed method was compared with other methods on both synthetic datasets and real-world datasets.The experimental results indicate that the proposed method achieves the top ranking on average across all 19 datasets while maintaining high precision and stability.

关 键 词：离群点检测 自适应邻居 面向边的方法 基于图的离群点检测 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论]