US 12,141,711 B2
Training or using sets of explainable machine-learning modeling algorithms for predicting timing of events
Jeffery Dugger, Atlanta, GA (US); and Michael McBurnett, Cumming, GA (US)
Assigned to Equifax Inc., Atlanta, GA (US)
Appl. No. 17/052,672
Filed by EQUIFAX INC., Atlanta, GA (US)
PCT Filed May 10, 2019, PCT No. PCT/US2019/031806
§ 371(c)(1), (2) Date Nov. 3, 2020,
PCT Pub. No. WO2019/217876, PCT Pub. Date Nov. 14, 2019.
Claims priority of provisional application 62/669,558, filed on May 10, 2018.
Prior Publication US 2021/0241141 A1, Aug. 5, 2021
Int. Cl. G06N 5/04 (2023.01); G06N 7/01 (2023.01); G06N 20/20 (2019.01); H04L 41/14 (2022.01); H04L 67/01 (2022.01)
CPC G06N 5/04 (2013.01) [G06N 7/01 (2023.01); G06N 20/20 (2019.01); H04L 41/14 (2013.01); H04L 67/01 (2022.05)] 20 Claims
OG exemplary drawing
 
1. A computing system comprising:
in a secured part of the computing system:
a data repository storing predictor data samples and response data samples, wherein (i) the predictor data samples include values of predictor variables that respectively correspond to actions performed by an entity or observations of the entity and (ii) each response data sample includes a respective outcome value of a response variable having a set of outcome values associated with the entity,
an external-facing subsystem configured for preventing a host server system from accessing the data repository via a data network, and
a development server system configured for:
accessing training data comprising a subset of the predictor data samples for a training window and a subset of the response data samples for the training window,
building a set of timing-prediction models from the training data, wherein building each timing-prediction model comprises training the timing-prediction model to predict a target event for a respective time bin within the training window, wherein the training window comprises two or more time bins and wherein at least two of the two or more time bins overlap,
generating program code configured to (i) compute a set of probabilities for the target event by applying the set of timing-prediction models to predictor variable data and (ii) compute a time of the target event from the set of probabilities, and
outputting the program code to the host server system via the external-facing subsystem; and
the host server system, wherein the host server system is communicatively coupled to the development server system and comprises one or more processing devices configured for executing the program code and thereby performing operations comprising:
receiving the predictor variable data,
computing a set of probabilities for the target event by applying the set of timing-prediction models to the predictor variable data,
computing a time of the target event from the set of probabilities, and
modifying a host system operation based on the computed time of the target event.