Weight adjusted composite model for forecasting in anomalous environments

• US 10,373,068 B2
• Filed: 11/10/2014
• Issued: 08/06/2019
• Est. Priority Date: 11/10/2014
• Status: Active Grant

• Current Assignee: Kyndryl Incorporated
• Original Assignee: International Business Machines Corporation
• Inventors: Baughman, Aaron K., Kozloski, James R., Mcavoy, Cameron N., O'Connell, Brian M.
• Primary Examiner(s): Waldron, Scott A.
• Assistant Examiner(s): Seck, Ababacar
• Application Number:
  US14/536,817
• Publication Number:
  US 20160132775A1
• Time in Patent Office: 1,730 Days
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 5/02 Knowledge representation; Symbolic representation

Abstract

A method, system, and computer program product for weight adjusted composite model for forecasting in anomalous environments are provided in the illustrative embodiments. A base forecasting model and a second forecasting model are combined to form a composite model, the base forecasting model configured to forecast an event in a time series, the second forecasting model configured to represent an anomalous portion of data in the time series. A mixing algorithm is combined with the composite model to adjust a set of weights associated with the composite model. Upon identifying a future period in which the event is to be forecasted, using the mixing algorithm, a subset of the set of weights is adjusted to from a weight adjusted composite model. The weight adjusted composite model is executed to forecast the event in the future period.

Claims

1. A method for event forecasting in anomalous environments, the method comprising:

in a forecasting data processing system comprising a software-implemented predictive model configured to cause a processor to perform a simulation of a phenomenon to forecast outcomes of the phenomenon at a future time,

creating new data of a composite forecasting model by combining with a base forecasting model a second forecasting model, the base forecasting model causing the processor to output a first forecast of an event in a time series, the data of the time-series comprising an anomalous portion, wherein the anomalous portion comprises a non-uniformity in a distribution of the event in the time-series, the second forecasting model being configured to represent the anomalous portion of data in the time series, wherein the second forecasting model comprises an equation whose curve fits, within a threshold, a curve formed by a set of values in the anomalous portion, and wherein the base forecasting model forecasts a value of the event in a non-anomalous portion of the data of the time series;

combining with the composite model a mixing algorithm, the mixing algorithm causing the processor to adjust a set of weights associated with the composite model;

adjusting, by the processor, responsive to identifying a future period in which the event is to be forecasted, using the mixing algorithm, existing weight data of a subset of the set of weights to produce new weight data of the subset of the set of weights, the processor using the new weight data to convert the composite forecasting model to a weight adjusted composite model in the event forecasting data processing system; and

executing, using the processor and a memory, the weight adjusted composite model to output a second forecast of the event during the future period in the anomalous portion of the time-series, wherein the second forecast has a better accuracy relative to the first forecast.

2. The method of claim 1, further comprising:

identifying a characteristic of the future period;

identifying a second period in a training data, wherein the characteristic of the future period matches a corresponding characteristic of the second period within a threshold; and

determining a portion of a curve representing a set of anomalous values in the training data during the second period, wherein the adjusting matches a curve representing the composite model during the future period with the portion of the curve representing the set of anomalous values in the training data during the second period.

3. The method of claim 2, wherein the training data comprises past data of the time series.

4. The method of claim 2, wherein the characteristic comprises a factor that affects the event during the future period.

5. The method of claim 2, wherein the characteristic comprises a time of day.

6. The method of claim 1, further comprising:

configuring the set of weights in the composite model, wherein a weight in the set of weights is associated with an expression in an equation in the composite model, wherein adjusting the weight causes a curve of the equation to change and form a changed curve, and wherein the changed fits, within a threshold, a second curve formed by a set of values in an expected anomalous portion in the time series during the future period.

7. The method of claim 1, wherein the second forecasting model comprises an equation representing a seasonal change in the time series.

8. The method of claim 1, wherein the second forecasting model comprises an equation representing a level change in the time series.

9. The method of claim 1, wherein the second forecasting model comprises an equation representing a trend change in the time series.

10. The method of claim 1, wherein the method is embodied in a computer program product comprising one or more computer-readable tangible storage devices and computer-readable program instructions which are stored on the one or more computer-readable tangible storage devices and executed by one or more processors.

11. The method of claim 1, wherein the method is embodied in a computer system comprising one or more processors, one or more computer-readable memories, one or more computer-readable tangible storage devices and program instructions which are stored on the one or more computer-readable tangible storage devices for execution by the one or more processors via the one or more memories and executed by the one or more processors.

12. A computer program product for event forecasting in anomalous environments, the computer program product comprising:

a computer readable storage medium;

program instructions of a forecasting data processing system comprising a software-implemented predictive model configured to cause a processor to perform a simulation of a phenomenon to forecast outcomes of the phenomenon at a future time;

program instructions, stored on the computer readable storage medium, to create new data of a composite forecasting model by combining with a base forecasting model a second forecasting model, the base forecasting model causing the processor to output a first forecast of an event in a time series, the data of the time-series comprising an anomalous portion, wherein the anomalous portion comprises a non-uniformity in a distribution of the event in the time-series, the second forecasting model being configured to represent the anomalous portion of data in the time series, wherein the second forecasting model comprises an equation whose curve fits, within a threshold, a curve formed by a set of values in the anomalous portion, and wherein the base forecasting model forecasts a value of the event in a non-anomalous portion of the data of the time series;

program instructions, stored on the computer readable storage medium, to combine with the composite model a mixing algorithm, the mixing algorithm causing the processor to adjust a set of weights associated with the composite model;

program instructions, stored on the computer readable storage medium, to adjust, by the processor, responsive to identifying a future period in which the event is to be forecasted, using the mixing algorithm, existing weight data of a subset of the set of weights to produce new weight data of the subset of the set of weights, the processor using the new weight data to convert the composite forecasting model to a weight adjusted composite model in the event forecasting data processing system; and

program instructions, stored on the computer readable storage medium, to execute, using the processor and a memory, the weight adjusted composite model to output a second forecast of the event during the future period in the anomalous portion of the time-series, wherein the second forecast has a better accuracy relative to the first forecast.

13. The computer program product of claim 12, further comprising:

program instructions, stored on the computer readable storage medium, to identify a characteristic of the future period;

program instructions, stored on the computer readable storage medium, to identify a second period in a training data, wherein the characteristic of the future period matches a corresponding characteristic of the second period within a threshold; and

program instructions, stored on the computer readable storage medium, to determine a portion of a curve representing a set of anomalous values in the training data during the second period, wherein the adjusting matches a curve representing the composite model during the future period with the portion of the curve representing the set of anomalous values in the training data during the second period.

14. The computer program product of claim 13, wherein the training data comprises past data of the time series.

15. The computer program product of claim 13, wherein the characteristic comprises a factor that affects the event during the future period.

16. The computer program product of claim 13, wherein the characteristic comprises a time of day.

17. The computer program product of claim 12, further comprising:

program instructions, stored on the computer readable storage medium, to configure the set of weights in the composite model, wherein a weight in the set of weights is associated with an expression in an equation in the composite model, wherein adjusting the weight causes a curve of the equation to change and form a changed curve, and wherein the changed fits, within a threshold, a second curve formed by a set of values in an expected anomalous portion in the time series during the future period.

18. A computer system for event forecasting in anomalous environments, the computer system comprising:

a processor, a memory, and a computer-readable storage device;

program instructions, stored on the computer-readable storage device for execution by the processor via the memory, the program instructions comprising;

a forecasting data processing system comprising a software-implemented predictive model configured to cause the processor to perform a simulation of a phenomenon to forecast outcomes of the phenomenon at a future time;

program instruction to create new data of a composite forecasting model by combining with a base forecasting model a second forecasting model, the base forecasting model causing the processor to output a first forecast of an event in a time series, the data of the time-series comprising an anomalous portion, wherein the anomalous portion comprises a non-uniformity in a distribution of the event in the time-series, the second forecasting model being configured to represent the anomalous portion of data in the time series, wherein the second forecasting model comprises an equation whose curve fits, within a threshold, a curve formed by a set of values in the anomalous portion, and wherein the base forecasting model forecasts a value of the event in a non-anomalous portion of the data of the time series;

program instruction to combine with the composite model a mixing algorithm, the mixing algorithm causing the processor to adjust a set of weights associated with the composite model;

program instruction to adjust, by the processor, responsive to identifying a future period in which the event is to be forecasted, using the mixing algorithm, existing weight data of a subset of the set of weights to produce new weight data of the subset of the set of weights, the processor using the new weight data to convert the composite forecasting model to a weight adjusted composite model in the event forecasting data processing system; and

program instruction to execute, using the processor and the memory, the weight adjusted composite model to output a second forecast of the event during the future period in the anomalous portion of the time-series, wherein the second forecast has a better accuracy relative to the first forecast.