US 12,437,310 B2
Method of training machine learning models for making simulated estimations
Alexander Krowitz, Biddeford, ME (US); and Martin Tapp, Montreal (CA)
Assigned to UKG Inc., Weston, FL (US)
Filed by Kronos Technology Systems Limited Partnership, Lowell, MA (US)
Filed on Sep. 19, 2022, as Appl. No. 17/933,422.
Application 17/933,422 is a continuation of application No. 17/036,050, filed on Sep. 29, 2020, granted, now 11,481,792.
Application 17/036,050 is a continuation of application No. 16/014,727, filed on Jun. 21, 2018, granted, now 11,068,916, issued on Jul. 20, 2021.
Claims priority of provisional application 62/524,792, filed on Jun. 26, 2017.
Prior Publication US 2023/0049931 A1, Feb. 16, 2023
This patent is subject to a terminal disclaimer.
Int. Cl. G06Q 30/02 (2023.01); G06N 5/01 (2023.01); G06N 7/00 (2023.01); G06N 20/00 (2019.01); G06N 20/20 (2019.01); G06Q 10/0631 (2023.01); G06Q 30/0202 (2023.01); G06N 3/02 (2006.01); G06N 5/02 (2023.01)
CPC G06Q 30/0202 (2013.01) [G06N 5/01 (2023.01); G06N 7/00 (2013.01); G06N 20/00 (2019.01); G06N 20/20 (2019.01); G06Q 10/063116 (2013.01); G06N 3/02 (2013.01); G06N 5/02 (2013.01)] 16 Claims
OG exemplary drawing
 
1. A method for training machine learning models for making simulated estimations, comprising:
collecting, via at least one processor from a database, historical data;
applying, via the at least one processor, one or more transformations to the historical data to create a set of model features;
separating, via the at least one processor, the set of model features into two or more pools, each pool comprising:
one or more model features of the set that are homogeneous with respect to a common value, and
a different type of feature from the other of the two or more pools;
for each of the two or more pools, creating, via the at least one processor, a corresponding training set comprising the one or more model features of the pool and at least some of the historical data;
for each of the two or more pools, creating, via the at least one processor, a corresponding machine learning model;
for each training set, training, via the at least one processor, the machine learning model that corresponds to the pool that corresponds to the training set, on the training set to predict a retail volume, wherein the machine learning model corresponding to the pool that corresponds to the training set is based at least in part on gradient boosting;
receiving, via the at least one processor, a request to generate a schedule for a time period;
determining, via the at least one processor, whether the requested time period includes at least one segment of time for which historical data is unavailable, and generating predicted historical retail volume data for the at least one segment of time using at least one of the machine learning models;
generating, via the at least one processor, the schedule based at least in part on the retail volume predicted by at least one of the machine learning models, wherein the generation of the schedule is based in part on the predicted historical retail volume data when the time period includes at least one segment of time for which historical data is unavailable;
updating the machine learning models via at least:
deleting and recreating the machine learning models, and
for each training set, retraining the machine learning model that corresponds to the pool that corresponds to the training set, on the training set to improve the prediction of the retail volume via improving a correlation of each model to the features within the corresponding pool;
wherein:
each of the set of model features corresponds to a store, and
the set of model features separated into the two or more pools comprises one or more special event features, each comprising a ratio between a retail volume on a date of a defined event and a retail volume forecasted for that date by the machine learning model without accounting for the defined event, wherein the one or more special event features are encoded using a scheme that preserves their ordering with respect to a business impact parameter.