基于核密度估计的分布数据流离群点检测.pdf

计算机研究与发展 IsSN1000一1239／CN11—1777／TP of Researchand JournalComputer Devdopment 基于核密度估计的分布数据流离群点检测 杨宜东1 孙志挥1 张 净1’2 1(东南大学计算机科学与工程系南京210096) 2(江苏大学电气信息工程学院镇江212001) (ydyang@seu．edu．cn) in DataStreams Kernel OutliersDistributed Basedon Finding Density Zhihuil，and YangYidon91，Sun ZhangJin91’2 1(忱抛以优删￡矿CD例吼￡Pr如心c已口，zdE咒gi撇；，Ii粥，S0“施∞盯‰iw坶勿，N妞砌!g210096) 212001) 2(GDZ如班盯上Ⅺ盯￡r0起站n，zd，行如rmn￡io咒E挖蓟H卵ri行g，J妇咒∥“U咒i粥耶i桫，ZhP力如珂g Abstract been stream therehas moreandmore basedondata models． Recently， occurring applications I)ata indata as ahotresearchfield．This mining stream，such clustering，classifying，etc，becomespaper an foroutlierdetectionindistributeddatastreams．Thedatastreamon presentsalgorithm every nodeistakenforasubsetofthe datastream，whichconsistsofdataonaUdistributed global of network traffic，itis tosendaHdatatoa centralnodeanddodetection．Basedonthe huge impossible communicationofdistributioninformationbetweendistributednodesandthecentral the node， algo“thm maintainsthe estimationfortheunionofaUstreams．Ondistributed outlierscan density every node，global bedetectedtheestimation．Detailsofcommunicationscheduleandoutlierdetectionarealsodiscussedin by this show results availabilitiesofthe paper．Experimentalpromising approach． wordsdistributeddata