US 12,298,963 B1
Optimized hampel filtering for outlier detection
Hongtao Hu, Cary, NC (US); and Mahesh V Joshi, Cary, NC (US)
Assigned to SAS Institute, Inc., Cary, NC (US)
Filed by SAS Institute Inc., Cary, NC (US)
Filed on Oct. 30, 2024, as Appl. No. 18/932,008.
Claims priority of provisional application 63/663,942, filed on Jun. 25, 2024.
Claims priority of provisional application 63/651,608, filed on May 24, 2024.
Claims priority of provisional application 63/647,948, filed on May 15, 2024.
Int. Cl. G06F 16/23 (2019.01); G06F 7/08 (2006.01); G06F 16/22 (2019.01); G06F 16/2455 (2019.01)
CPC G06F 16/2365 (2019.01) [G06F 7/08 (2013.01); G06F 16/22 (2019.01); G06F 16/24558 (2019.01)] 30 Claims
OG exemplary drawing
 
1. A computer-program product tangibly embodied in a non-transitory machine-readable storage medium, the computer-program product including instructions operable to cause a computing device to perform operations, comprising:
accessing a memory storing a data structure, the data structure comprising a set of sorted values sampled from a plurality of values of a dataset;
writing a new value from the plurality of values to the data structure, wherein the new value replaces an oldest value of the set of sorted values, and wherein the new value is inserted in a sorted position within the set of sorted values different than an original position of the oldest value;
modifying the data structure by subtracting a median value of the set of sorted values from each value of the set of sorted values to obtain a corresponding set of sorted signed deviation values;
segmenting the set of sorted signed deviation values of the data structure at a segmentation point to obtain a first data substructure comprising a first subset of sorted absolute deviation values and a second data substructure comprising a second subset of sorted absolute deviation values, wherein the segmentation point is positioned between a negative signed value and a non-negative value within the set of sorted signed deviation values;
performing a binary search on the first data substructure and the second data substructure to identify a median absolute deviation value;
computing a difference between a particular value of the set of sorted values and the median value of the set of sorted values, wherein the particular value comprises a value from the dataset being evaluated for an outlier decision; and
based on whether the difference is less than a threshold value computed from the median absolute deviation value, generating an outlier decision output indicative of whether the particular value of the set of sorted values comprises an outlier value.