一种改进的离群点检测方法.PDFVIP

35 3 ( ) V o.l 35 No. 3 2007 6 Journa l of F uzhou U n iversity( N a tura l Sc ience) Jun. 2007 1000 - 224 3( 2007) 03- 0376- 05 , (, 350002) , . , k th , k , . DokOF . , . ; ; D okOF TP 39 A An improved efficient algorithm for detecting outliers CHEN Guang- p ing, Y E D ong- y i ( College o f M athem atics and Com puter Science, Fuzhou U niv ers ity, F uzhou, F uj ian 350002, Ch ina) Abstract: Ex isting algo rithm s for outlier detect ion are no t effic ient when facing large datasets. W ith analysis of features o f out liers in datasets, w e rank each point on the basis o f its d istance to its k th nea rest ne ighbo r and the distribution o f itsk nearest neighbo rs. A nove l algorithm, ca lled DokOF ( outlier factor b ased on distribution o fk nearest ne ighbors), is presented, wh ich can dea l w ith m ixed attribute data. R esults of expermi ents dem onstra te prom ising availab ilities of th is approach. Keywords: da ta m in ing; outlier detection; DokOF algor ithm 1 [ 1] . H awk in s , . , . , . . [ 2] B reun