US 2017/0364614 A1Published application

Adaptive forecasting of time-series

AssigneeInternational Business Machines Corporation

InventorsMARK A. FREEMAN, Michael E. Girata, Mark R. Grabau

Life of this patent

Estimating the term…

Loading patent…

Abstract

For a time-series, a baseline error value is reduced to compute a target forecast error value by maximizing a net benefit value of a forecasted value of the time-series. For each forecasting model in a set of models, a corresponding model error value related to the time-series is computed. From the set, a subset of models is selected where each model in the subset has a cost that will produce a positive value for the net benefit. A selected model from the subset is associated with the time-series such that a model error value of the selected model is at most equal to the target forecast error. The time-series is forecasted using the selected model such that the forecasted value has an error of less than the baseline error at a future time, and the forecasted value produces a positive net benefit at the future time.

Claims20

Dependent claims

A method comprising:

reducing, using a processor and a memory, for a time-series, a baseline error value to compute a target forecast error value, wherein the reducing maximizes a net benefit value of a forecasted value of the time-series;
computing, for each forecasting model in a set of forecasting models, a corresponding model error value, wherein the model error value is related to the time-series;
selecting from the set of forecasting models, a subset of forecasting models, wherein each member model of the subset has a cost that will produce a positive value for the net benefit value;
associating with the time-series, a selected model from the subset such that a model error value of the selected model is at most equal to the target forecast error value;
and
forecasting the time-series using the selected model, wherein the forecasting predicts the forecasted value of the time-series at a future time such that the forecasted value has an error of less than the baseline error value relative to an actual value of the time-series at the future time, and such that the forecasted value produces a positive net benefit value at the future time.

3 more elements

A computer usable program product comprising one or more computer-readable storage devices, and program instructions stored on at least one of the one or more storage devices, the stored program instructions comprising:

program instructions to reduce, using a processor and a memory, for a time-series, a baseline error value to compute a target forecast error value, wherein the reducing maximizes a net benefit value of a forecasted value of the time-series;
program instructions to compute, for each forecasting model in a set of forecasting models, a corresponding model error value, wherein the model error value is related to the time-series;
program instructions to select from the set of forecasting models, a subset of forecasting models, wherein each member model of the subset has a cost that will produce a positive value for the net benefit value;
program instructions to associate with the time-series, a selected model from the subset such that a model error value of the selected model is at most equal to the target forecast error value;
and
program instructions to forecast the time-series using the selected model, wherein the forecasting predicts the forecasted value of the time-series at a future time such that the forecasted value has an error of less than the baseline error value relative to an actual value of the time-series at the future time, and such that the forecasted value produces a positive net benefit value at the future time.

3 more elements

A computer system comprising one or more processors, one or more computer-readable memories, and one or more computer-readable storage devices, and program instructions stored on at least one of the one or more storage devices for execution by at least one of the one or more processors via at least one of the one or more memories, the stored program instructions comprising:

program instructions to reduce, for a time-series, a baseline error value to compute a target forecast error value, wherein the reducing maximizes a net benefit value of a forecasted value of the time-series;
program instructions to compute, for each forecasting model in a set of forecasting models, a corresponding model error value, wherein the model error value is related to the time-series;
program instructions to select from the set of forecasting models, a subset of forecasting models, wherein each member model of the subset has a cost that will produce a positive value for the net benefit value;
program instructions to associate with the time-series, a selected model from the subset such that a model error value of the selected model is at most equal to the target forecast error value;
and
program instructions to forecast the time-series using the selected model, wherein the forecasting predicts the forecasted value of the time-series at a future time such that the forecasted value has an error of less than the baseline error value relative to an actual value of the time-series at the future time, and such that the forecasted value produces a positive net benefit value at the future time.

3 more elements

Dependencies are inferred from the claim text.