detect anomal record categor dataset kaustav da jeff schneider problem detect anomali high aritycategor dataset applic anomali defin datapoint abnorm access data consist normal record long small percentag unlabel anomal record interest problem unsupervis anomali detect unlabel data train detect record follow definit normal standard approach creat model normal data compar test record probabilist approach build likelihood model train data record test anomali base complet record likelihood probabl model categor attribut bay net give standard represent likelihood approach good find outlier dataset detect record attribut valu rare detect rare valu attribut desir outlier consid anomali context present altern definit anomali propos approach compar margin distribut attribut subset show meaning detect anomali perform semi synthet real world dataset 