三种不同监督方法的离群值检测在欺诈交易上的比较

Comparison of Outlier Detection of Three Different Supervised Methods in Fraudulent Transactions

离群值检测的目的在于检测出不符合期望行为的数据,因而适合应用于如欺诈交易检测等多个领域。针对目前众多的通用离群值检测方法,本文侧重从机器学习的角度出发,分别从基于无监督、监督和半监督技术分析与比较三种不同销售欺诈交易检测模型,使用决策精确/回溯精确度和累积回溯精确度对其不同实验结果进行分析比较。

Outlier detection aims to detect those data that significantly deviate from the expected behavior, and thus is widely applied in fraudulent transaction detection, etc. In view of many algorithms for outlier detection, this paper, starting from machine learning, makes an analysis and comparison on three different fraudulent sale transaction models based respectively on unsupervised, supervised and semisupervised technologies, and it analyzes and compares their experiment results by precision/recall （PR） and cumulative recall（CR）.