US 12,367,399 B2
Temporally dynamic location-based predictive data analysis
Alison R. Stroh, Minnetonka, MN (US); Mario M. Suarez, Ashburn, VA (US); Stephen R. Dion, Reading, MA (US); Jordan R. DiPascal, Knightdale, NC (US); and Derek J. Syverson, Minneapolis, MN (US)
Assigned to UnitedHealth Group Incorporated, Minnetonka, MN (US)
Filed by UnitedHealth Group, Incorporated, Minnetonka, MN (US)
Filed on Jun. 1, 2021, as Appl. No. 17/335,260.
Claims priority of provisional application 63/085,226, filed on Sep. 30, 2020.
Prior Publication US 2022/0101150 A1, Mar. 31, 2022
Int. Cl. G06N 5/02 (2023.01); G06N 20/00 (2019.01)
CPC G06N 5/02 (2013.01) [G06N 20/00 (2019.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method comprising:
extracting, by one or more processors and from an external data source, (1) an input locality data object, indicating a geographic area, and (2) an input control policy data object, indicating an event in the geographic area, wherein the input locality data object comprises an input locality sentiment designation and an input locality adherence designation associated with the input control policy data object;
generating, by the one or more processors and by using a cohort generation clustering machine learning model, a locality cohort, comprising a set of one or more cohort locality data objects from a plurality of cohort locality data objects, for the input locality data object and the input control policy data object by processing a mapping of the plurality of cohort locality data objects to a multi-dimensional mapping space, the multi-dimensional mapping space comprising a group of mapping dimensions that comprises a first mapping dimension associated with a plurality of locality sentiment designations and a second mapping dimension associated with a plurality of locality adherence designations, wherein a cohort locality data object of the set of one or more cohort locality data objects comprises an indication of (i) the input control policy data object, (ii) a sentiment designation that corresponds to the input locality sentiment designation, and (iii) an adherence designation that corresponds to the input locality adherence designation;
generating, by the one or more processors and by using a cohort-based growth forecast recurrent neural network machine learning model, an inferred cross-temporal growth prediction for the input locality data object with respect to the input control policy data object and in relation to a plurality of policy-indexed temporal units, wherein the cohort-based growth forecast recurrent neural network machine learning model is configured to:
(i) identify a predecessor subset for a policy-indexed temporal unit of the plurality of policy-indexed temporal units that comprises the policy-indexed temporal unit and one or more of the plurality of policy-indexed temporal units that temporally precede the policy-indexed temporal unit,
(ii) process a plurality of ground truth cross-temporal growth data objects associated with one or more of the plurality of cohort locality data objects that is in the predecessor subset to generate a plurality of inferred temporal growth predictions associated with the plurality of policy-indexed temporal units, and
(iii) determine the inferred cross-temporal growth prediction based at least in part on the plurality of inferred temporal growth predictions; and
providing, by the one or more processors and to a client computing entity, a predicted growth trend for the input locality data object using a graph-based user interface element and based at least in part on the inferred cross-temporal growth prediction.