US 11,886,516 B1Granted

Generating synthetic data based on time series predictions and plural machine learning models

AssigneeCapital One Services, LLC

InventorsAustin Walters, Jeremy Goodsitt

Life of this patent

Estimating the term…

Loading patent…

Abstract

Aspects described herein may relate to methods, systems, and apparatuses for generating synthetic data based on time series predictions and plural machine learning models. Generating the synthetic data may include receiving one or more data records that include transaction data of a user, categorizing and/or segmenting the one or more data records, determining a first predicted time step based on the categorizing and/or segmenting, and performing an iterative process that determines further predicted time steps. The first predicted time step may be determined using one model of a plurality of machine learning models. The iterative process may determine the further predicted time steps using the plural machine learning models. Based on iterations of the iterative process, potential time series may be determined. The synthetic time series may be determined from the potential time series. The synthetic time series may be used by additional processes, such as additional machine-learning processes.

Claims20

Dependent claims

A method comprising:

receiving, by one or more computing devices, one or more data records that include transaction data of a user;
based on applying one or more classification techniques to the one or more data records, determining, by the one or more computing devices, a first category for the one or more data records, wherein the first category is of a plurality of categories;
determining, by the one or more computing devices and based on the one or more data records, data record segments, wherein each segment of the data record segments includes a time step of the transaction data;
based on the data record segments, determining, by the one or more computing devices, first time series model input;
using a first model and the first time series model input, determining, by the one or more computing devices, a first predicted time step of predicted transaction data, wherein the first predicted time step is associated with the first category, wherein the first model is configured to predict first time steps associated with the first category, and wherein the first model is one of a plurality of machine learning models configured to predict time steps associated with the plurality of categories;
determining, by the one or more computing devices, second time series model input that includes the first predicted time step;
using the plurality of machine learning models and the second time series model input, determining, by the one or more computing devices, second predicted time steps of predicted transaction data, wherein the second predicted time steps are associated with the plurality of categories;
based on the first predicted time step and the second predicted time steps, determining, by the one or more computing devices, a plurality of potential time series of predicted transaction data;
determining, by the one or more computing devices, confidence values for the plurality of potential time series;
and
based on the confidence values and based on the plurality of potential time series, determining, by the one or more computing devices, a synthetic time series for the user, wherein the synthetic time series includes the first predicted time step and one of the second predicted time steps.

8 more elements

An apparatus comprising:

one or more processors;
and
memory storing executable instructions that, when executed by the one or more processors, cause the apparatus to:
receive one or more data records that include transaction data of a user;
based on applying one or more classification techniques to the one or more data records, determine a first category for the one or more data records, wherein the first category is of a plurality of categories;
determine, based on the one or more data records, data record segments, wherein each segment of the data record segments includes a time step of the transaction data;
based on the data record segments, determine first time series model input;
using a first model and the first time series model input, determine a first predicted time step of predicted transaction data, wherein the first predicted time step is associated with the first category, wherein the first model is configured to predict first time steps associated with the first category, and wherein the first model is one of a plurality of machine learning models configured to predict time steps associated with the plurality of categories;
determine second time series model input that includes the first predicted time step;
using the plurality of machine learning models and the second time series model input, determine second predicted time steps of predicted transaction data, wherein the second predicted time steps are associated with the plurality of categories;
determine, based on the first predicted time step and the second predicted time steps, a plurality of potential time series of predicted transaction data;
determine, by the one or more computing devices, confidence values for the plurality of potential time series;
and
based on the confidence values and based on the plurality of potential time series, determine a synthetic time series for the user, wherein the synthetic time series includes the first predicted time step and one of the second predicted time steps.

11 more elements

One or more non-transitory computer-readable media storing executable instructions that, when executed, cause one or more computing devices to:

receive one or more data records that include transaction data of a user;
based on applying one or more classification techniques to the one or more data records, determine a first category for the one or more data records, wherein the first category is of a plurality of categories;
determine, based on the one or more data records, data record segments, wherein each segment of the data record segments includes a time step of the transaction data;
based on the data record segments, determine first time series model input;
using a first model and the first time series model input, determine a first predicted time step of predicted transaction data, wherein the first predicted time step is associated with the first category, wherein the first model is configured to predict first time steps associated with the first category, and wherein the first model is one of a plurality of machine learning models configured to predict time steps associated with the plurality of categories;
determine second time series model input that includes the first predicted time step;
using the plurality of machine learning models and the second time series model input, determine second predicted time steps of predicted transaction data, wherein the second predicted time steps are associated with the plurality of categories;
determine, based on the first predicted time step and the second predicted time steps, a plurality of potential time series of predicted transaction data;
determine, by the one or more computing devices, confidence values for the plurality of potential time series;
and
based on the confidence values and based on the plurality of potential time series, determine a synthetic time series for the user, wherein the synthetic time series includes the first predicted time step and one of the second predicted time steps.

8 more elements

Dependencies are inferred from the claim text.