angl base outlier detect high dimension data han peter kriegel matthia hubert arthur zimek detect outlier larg set data object major data mine task aim find mechan respons group object data set exist approach base assess distanc indirectli assum distribut full dimension euclidean data space high dimension data approach bound deterior due notori curs dimension paper propos approach name abod angl base outlier detect variant assess varianc angl differ vector point point effect curs dimension allevi compar pure distanc base approach main advantag approach method reli paramet select influenc qualiti achiev rank experiment evalu compar abod establish distanc base method lof artifici real world data set show abod perform high dimension data 