# US 11,443,015 B2 · Generating prediction models in accordance with any specific data sets

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 11,443,015 B2
- Document ID: US11443015B2
- Status: Granted
- Kind: B2
- Application number: 14/919,395
- Filed: 2015-10-21
- Published: 2017-04-27
- Granted: 2022-09-13
- Earliest claimed: 2015-10-21
- Assignees: Adobe Inc.
- Applicants: Adobe Inc.
- Inventors: Kourosh Modarresi
- Web page: https://patentgrep.com/patent/US11443015B2

## CPC

- G06F 17/18: for evaluating statistical data {, e.g. average values, frequency distributions, probability functions, regression analysis (forecasting specially adapted for a specific administrative, business or logistic context G06Q10/04)}
- G06N 20/00: Machine learning
- G06Q 10/04: Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem" (market predictions or forecasting for commercial activities G06Q30/0202)
- G06Q 10/063: Operations research, analysis or management
- G06Q 10/067: Enterprise or organisation modelling
- G06Q 30/02: Marketing; Price estimation or determination; Fundraising
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06Q: INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR

## Abstract

Embodiments of the present invention provide systems, methods, and computer storage media for facilitating generation of prediction models. In some embodiments, a predetermined number of parameter value sets is identified. Each parameter value set includes a plurality of parameter values that represent corresponding parameters within a time series model. The parameter values can be selected in accordance with stratified sampling to increase a likelihood of prediction accuracy. The parameter value sets are input into a time series model to generate a prediction value in accordance with observed time series data, and the parameter value set resulting in a least amount of prediction error can be selected and used to generate a time series prediction model (ARIMA, AR, MA, ARMA) with corresponding model parameters, such as p, q, and/or k, subsequently used to predict values.

## Claims

### Claim 1

One or more non-transitory computer storage media storing computer-useable instructions that, when used by one or more computing devices, cause the one or more computing devices to perform operations to facilitate generation of prediction models, the method comprising:
identifying a predetermined number of parameter value sets, each parameter value set having parameter values that represent corresponding parameters for inputting as a combination to define a function in an autoregressive integrated moving average (ARIMA) prediction model, wherein the parameter values indicate one of a number of differencing terms or an order of differencing terms to use in the corresponding function of the ARIMA predication model, and an indication of a previous number of web page views, each parameter value within each parameter value set is selected by:
selecting a value from one of a plurality of strata, for a corresponding parameter, based on probabilities associated with each strata of the plurality of strata for the corresponding parameter, wherein each strata of the plurality of strata for the corresponding parameter includes a subportion of a plurality of potential parameter values for the corresponding parameter and the strata is associated with a corresponding probability, used to select the value, determined based on a distribution of previously selected parameter values for previous prediction models;
receiving a first portion of the observed time series data set from a data collection center with data describing web traffic;
in a first iteration, implementing each identified parameter value set into the corresponding ARIMA prediction model and use the first portion of the observed time series data set to generate a predicted value associated with each parameter value set;
comparing the predicted value associated with each parameter value set to a second portion of the observed time series data set that corresponds with the predicted value to generate a first relative prediction error for each parameter value set, wherein the first portion does not overlap with the second portion;
in a second iteration, implementing each identified parameter value set into the corresponding ARIMA prediction model and use a third portion of the observed time series data set to generate a new predicted value associated with each parameter value set, wherein the third portion partially overlaps with the second portion;
comparing the new predicted value associated with each parameter value set to a fourth portion of the observed time series data set that corresponds with the predicted value to generate a second relative prediction error for each parameter value set, wherein the fourth portion does not overlap with the third portion;
selecting a parameter value set associated with a least relative prediction error, from the first and the second relative prediction errors for each parameter value set, for generating a time series prediction model;
implementing the selected parameter value set into the ARIMA prediction model to generate the time series prediction model; and
using the time series prediction model to predict values expected to occur at a later time.

### Claim 2 (depends on 1)

The one or more computer storage media of claim 1, wherein each parameter value within each parameter value set is selected in accordance with a given range of parameter values for each parameter, the given range including the plurality of potential parameter values for the corresponding parameter selected based on recent time series data of the observed time series data set, wherein a first given range for an order of differences is less than a second given range for the previous number of web page views.

### Claim 3 (depends on 1)

The one or more computer storage media of claim 1, wherein the generation of the time series prediction model is initiated based on a request for one or more prediction values.

### Claim 4 (depends on 1)

The one or more computer storage media of claim 1, wherein the parameters within the time series prediction model further comprise a number of differencing terms and a number of lagged white noise terms.

### Claim 5 (depends on 1)

The one or more computer storage media of claim 1, wherein the time series prediction model comprises an autoregressive integrated moving average (ARIMA) model.

### Claim 6 (depends on 1)

The one or more computer storage media of claim 1, the method further comprising:
selecting a portion of the plurality of potential parameter values based on a predetermined number of recent observed time series data values of a set of observed time series data;
for each parameter, dividing the portion of the plurality of potential parameter values into the plurality of strata, wherein each strata of the plurality of strata includes the subportion of the portion of the plurality of potential parameter values; and
assigning the corresponding probability to each strata.

### Claim 7 (depends on 1)

The one or more computer storage media of claim 1, wherein selecting the parameter value set associated with the least relative of prediction error comprises comparing the prediction value with a corresponding observed time series data.

### Claim 8 (depends on 1)

The one or more computer storage media of claim 1, wherein the ARIMA predication model utilizes each of p, q, and k parameters.

### Claim 9 (depends on 8)

The one or more computer storage media of claim 8, wherein the q parameter indicates a number of lagged white noise terms.

### Claim 10

A method to facilitate generation of prediction models, the method comprising:
identifying a predetermined number of parameter value sets, each parameter value set having parameter values that represent corresponding parameters for inputting as a combination to define a function in an autoregressive integrated moving average (ARIMA) prediction model, wherein the parameter values indicate one of a number of differencing terms or an order of differencing terms to use in the corresponding function of the ARIMA predication model, and an indication of a previous number of web page views, each parameter value within each parameter value set is selected by:
selecting a value from one of a plurality of strata, for a corresponding parameter, based on probabilities associated with each strata of the plurality of strata for the corresponding parameter, wherein each strata of the plurality of strata for the corresponding parameter includes a subportion of a plurality of potential parameter values for the corresponding parameter and the strata is associated with a corresponding probability, used to select the value, determined based on a distribution of previously selected parameter values for previous prediction models;
receiving a first portion of the observed time series data set from a data collection center with data describing web traffic;
in a first iteration, implementing each identified parameter value set into the corresponding ARIMA prediction model and use the first portion of the observed time series data set to generate a predicted value associated with each parameter value set;
comparing the predicted value associated with each parameter value set to a second portion of the observed time series data set that corresponds with the predicted value to generate a first relative prediction error for each parameter value set, wherein the first portion does not overlap with the second portion;
in a second iteration, implementing each identified parameter value set into the corresponding ARIMA prediction model and use a third portion of the observed time series data set to generate a new predicted value associated with each parameter value set, wherein the third portion partially overlaps with the second portion;
comparing the new predicted value associated with each parameter value set to a fourth portion of the observed time series data set that corresponds with the predicted value to generate a second relative prediction error for each parameter value set, wherein the fourth portion does not overlap with the third portion;
selecting a parameter value set associated with a least relative prediction error, from the first and the second relative prediction errors for each parameter value set, for generating a time series prediction model;
implementing the selected parameter value set into the ARIMA prediction model to generate the time series prediction model; and
using the time series prediction model to predict values expected to occur at a later time.

### Claim 11 (depends on 10)

The method of claim 10, wherein each parameter value within each parameter value set is selected in accordance with a given range of parameter values for each parameter, the given range of parameter values including the plurality of potential parameter values of the corresponding parameter, the plurality of potential parameter values selected based on a distribution of previous parameter values associated with one or more of previous prediction models.

### Claim 12 (depends on 10)

The method of claim 10, wherein the time series prediction model is performed in real-time.

### Claim 13 (depends on 10)

The method of claim 10, wherein the time series prediction model comprises the ARIMA prediction model or autoregressive moving average model (ARMA).

### Claim 14 (depends on 10)

The method of claim 10, wherein the time series prediction model comprises a moving average model.

### Claim 15 (depends on 10)

The method of claim 10, wherein selecting the parameter value set associated with the least relative amount of prediction error comprises comparing the prediction value with a corresponding observed time series data.

### Claim 16

A system comprising:
one or more processors; and
one or more computer storage media storing computer-useable instructions that, when used by the one or more processors, cause the one or more processors to:
identify a predetermined number of parameter value sets, each parameter value set having parameter values that represent corresponding parameters for inputting as a combination to define a function in an autoregressive integrated moving average (ARIMA) prediction model, wherein the parameter values indicate one of a number of differencing terms or an order of differencing terms to use in the corresponding function of the ARIMA predication model, and an indication of a previous number of web page views, each parameter value within each parameter value set is selected by:
selecting a value from one of a plurality of strata, for a corresponding parameter, based on probabilities associated with each strata of the plurality of strata for the corresponding parameter, wherein each strata of the plurality of strata for the corresponding parameter includes a subportion of a plurality of potential parameter values for the corresponding parameter and the strata is associated with a corresponding probability, used to select the value, determined based on a distribution of previously selected parameter values for previous prediction models;
receiving a first portion of the observed time series data set from a data collection center with data describing web traffic;
in a first iteration, implement each identified parameter value set into the corresponding ARIMA prediction model and use the first portion of the observed time series data set to generate a predicted value associated with each parameter value set;
compare the predicted value associated with each parameter value set to a second portion of the observed time series data set that corresponds with the predicted value to generate a first relative prediction error for each parameter value set, wherein the first portion does not overlap with the second portion;
in a second iteration, implement each identified parameter value set into the corresponding ARIMA prediction model and use a third portion of the observed time series data set to generate a new predicted value associated with each parameter value set, wherein the third portion partially overlaps with the second portion;
compare the new predicted value associated with each parameter value set to a fourth portion of the observed time series data set that corresponds with the predicted value to generate a second relative prediction error for each parameter value set, wherein the fourth portion does not overlap with the third portion;
select a parameter value set associated with a least relative prediction error, from the first and the second relative prediction errors for each parameter value set, for generating a time series prediction model;
implement the selected parameter value set into the ARIMA prediction model to generate the time series prediction model; and
use the time series prediction model to predict values expected to occur at a later time.

### Claim 17 (depends on 16)

The system of claim 16, wherein the time series prediction model is performed in real-time.

### Claim 18 (depends on 16)

The system of claim 16, wherein the time series prediction model comprises the ARIMA prediction model or autoregressive moving average model (ARMA).

### Claim 19 (depends on 16)

The system of claim 16, wherein the time series prediction model comprises a moving average model.

### Claim 20 (depends on 16)

The system of claim 16, wherein the time series prediction model comprises an autoregressive model.