Predictive modeling across multiple horizons combining time series and external data
A multi-horizon predictor system that predicts a future parameter value for multiple horizons based on time-series data of the parameter, external data, and machine-learning. For a given time horizon, a time series data splitter splits the time into training data corresponding to a training time period, and a validation time period corresponding to a validation time period between the training time period and the given horizon. A model tuner tunes the prediction model of the given horizon fitting an initial prediction model to the parameter using the training data thereby using machine learning. The model tuner also tunes the initial prediction model by adjusting an effect of the external data on the prediction to generate a final prediction model for the given horizon using the validation data. A multi-horizon predictor causes the time series data splitter and the model tuner to operate for each of multiple horizons.
1. A computing system that automatically predicts a future value of a parameter for multiple horizons, each horizon comprising a future term comprising a period of time, based on time-series data of the parameter, external data, and machine-learning, the computing system comprising:
a time series data splitter that, for a given horizon, splits time series data of a parameter into training data corresponding to a training time period prior to the given horizon, and a validation time period corresponding to a validation time period between the training time period and the given horizon;
a model tuner that tunes a prediction model of the given horizon by performing the following for the given horizon:
fitting an initial prediction model to the parameter using the training data, thereby using machine learning;
tuning the initial prediction model by adjusting an effect of the external data on the prediction, wherein the external data comprises data that are distinct from the time series data and the parameter; and
using the adjusted initial prediction model to generate a final prediction model for the given horizon using the validation data, such that the prediction accounts for the external data; and
a multi-horizon predictor that causes the time series data splitter and the model tuner to operate for each a plurality of horizons, using both a different tuning and a corresponding different final prediction model for each of at least some of the plurality of horizons, to generate a single multi-horizon prediction result comprising predictions for multiple horizons where each horizon comprises a future term comprising a period of time.
2. The system in accordance with claim 1 , the multi-horizon predictor keeping the same initial prediction model for each of the plurality of horizons.
3. The system in accordance with claim 1 , the multi-horizon predictor changing the initial prediction model for at least one of the plurality of horizons.
4. The system in accordance with claim 1 , the external data comprising data originating remotely from the computing system.
5. The system in accordance with claim 1 , the external data comprising changing external data, the system further comprising:
an external data change detection component that causes the multi-horizon predictor to redo the multi-horizon prediction in response to at least some changes in the external data.
6. The system in accordance with claim 1 , the multi-horizon prediction model periodically redoing the multi-horizon prediction based on shifted time series data produced by the time series data splitter.
7. The system in accordance with claim 6 , the multi-horizon prediction model periodically redoing the multi-horizon prediction also based on updated external data.
8. The system in accordance with claim 1 , the parameter comprising a resource usage parameter.
9. The system in accordance with claim 1 , the parameter comprising a product flow parameter.
10. The system in accordance with claim 1 , the external data comprising a prediction value also predicted using a time series.
11. The system in accordance with claim 1 , the external data comprising macroeconomic data.
12. The system in accordance with claim 1 , the external data comprising at least two different types of data.
13. The system in accordance with claim 1 , the external data comprising a scheduled event.
14. The system in accordance with claim 1 , the external data comprising weather forecast data.
15. The system in accordance with claim 1 , the external data comprising search engine query data.
16. The system in accordance with claim 1 , the external data comprising customer opportunity data.
17. The system in accordance with claim 1 , the parameter being specific to a geographic region.
18. The system in accordance with claim 1 , the external data comprising data regarding the geographic region.
19. A method for automatically predicting a future value of a parameter for multiple horizons based on time-series data of the parameter, external data comprising data that are distinct from the time-series data and the parameter, and machine-learning, the method comprising:
for each of multiple horizons, each horizon comprising a future term comprising a period of time, performing the following to generate a single multi-horizon prediction result comprising predictions for multiple horizons:
splitting time series data of a parameter, for a given horizon, into training data corresponding to a training time period prior to the given horizon, and a validation time period corresponding to a validation time period between the training time period and the given horizon; and
tuning a prediction model of the given period by performing the following for the given horizon:
fitting an initial prediction model to the parameter using the training data thereby using machine learning; and
tuning the initial prediction model by adjusting an effect of the external data on the prediction to generate a final prediction model for the given horizon using the validation data, the adjusting allowing for a prediction that accounts for the external data.
20. A computer program product comprising one or more computer-readable hardware storage media having thereon computer-executable instructions that are structured such that, when executed by one or more processors of a computing system, cause the computing system to perform a method for automatically predicting a future value of a parameter for multiple horizons based on time-series data of the parameter, external data, and machine-learning, the method comprising:
for each of multiple horizons, each horizon comprising a future term comprising a period of time, performing the following to generate a single multi-horizon prediction result comprising predictions for multiple horizons:
splitting time series data of a parameter, for a given horizon, into training data corresponding to a training time period prior to the given horizon, and a validation time period corresponding to a validation time period between the training time period and the given horizon; and
tuning a prediction model of the given period by performing the following for the given horizon:
fitting an initial prediction model to the parameter using the training data thereby using machine learning; and
tuning the initial prediction model by adjusting an effect of the external data on the prediction to generate a final prediction model for the given horizon using the validation data, the adjusting allowing for a prediction that accounts for the external data.