Knorr, E. M., and Ng, R. T. 1997. A unified notion of outliers. Unpublished Manuscript, Dept. of Computer Science, University of British Columbia.  Home/Search   Document Not in Database   Summary   Related Articles

....the normal curve. Nevertheless, the same proof procedure shown above still works with appropriate modifications. As a concrete example, Barnett and Lewis give DefN6 for testing for an upper and lower outlier pair in a normal distribution with unknown mean and variance. In a more detailed report (Knorr Ng 1997), we give a proof showing that UO(p; D) unifies DefN6 with p 0 = 0:999 and D 0 = 0:2. Due to very limited space, we omit those details. Details and proofs are also available for the following lemmas. Outliers in Other Distributions Consider the following test for finding a single upper outlier ....

....to k 2, we use the same algorithm; however, to maintain Properties 1 4, i) the length of a cell changes from l = D 2 p 2 to l = D 2 p k , and (ii) Layer 2 is no longer 2 cells thick, but rather d2 p ke cells thick. We defer proofs and complexity analysis to a more detailed report (Knorr Ng 1997). In ongoing work, we are studying how to optimize UO(p; D) outlier detection for high dimensional cases, e.g. k 10. Preliminary Experimental Results We ran our algorithm on a 16 attribute, 856 record dataset consisting of 1995 96 performance statistics for players in the National Hockey ....

Knorr, E. M., and Ng, R. T. 1997. A unified notion of outliers. Unpublished Manuscript, Dept. of Computer Science, University of British Columbia.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC