Systems and methods for generating optimal data predictions in real-time for time series data signals
Methods and systems are disclosed for generating optimal data predictions in time series data signals based on empirically-optimized model selection, noise filtering, and window size selection using machine learning models. For example, the system may receive a first subset of time series data. The system may receive a prediction horizon. The system may generate a feature input based on the first subset of time series data and the prediction horizon. The system may input the feature input into a machine learning model, wherein the machine learning model includes multiple components. The system may receive an output from the machine learning model. The system may generate for display, on a user interface, a prediction for the first subset of time series data at the prediction horizon based on the output.
1. A system for generating optimal data predictions in real-time for time series data signals related to load processing in disparate computer networks based on empirically-optimized model selection, noise filtering, and window size selection using machine learning models, the system comprising:
cloud-based storage circuitry configured to store:
an ensemble of base models; and
a machine learning model, wherein the machine learning model includes:
a first model component, wherein the first model component is trained to select, for given time series data and given prediction horizons, an optimal base model collection for the ensemble of base models and select an optimal parameter set for the optimal base model collection, and wherein the optimal base model collection and optimal parameter set are selected by evaluating multiple candidate model structures and associated parameters to minimize prediction error and maximize a predictive performance metric;
a second model component, wherein the second model component is trained to select a filtering parameter for the given time series data and given prediction horizons; and
a third model component, wherein the third model component is trained to select an optimal window size for the given time series data and given prediction horizons;
cloud-based control circuitry configured to:
receive a first subset of time series data, wherein the first subset of times series data comprises values for load amounts for processing requests on computer nodes;
receive a prediction horizon, wherein the prediction horizon comprises a future date for predicting a load amount for processing requests;
generate a feature input based on the first subset of time series data and the prediction horizon, wherein the feature input comprises a vectorized representation of the first subset of time series data and the prediction horizon;
input the feature input into the machine learning model; and
receive an output from the machine learning model; and
cloud-based input/output circuitry configured to generate for display, on a user interface, a prediction for the first subset of time series data at the prediction horizon based on the output, wherein the prediction comprises a graphical representation of a predicted set of values for time series values and a confidence level for each of the predicted set of values for time series values.
2. A method for generating optimal data predictions in real-time for time series data signals based on empirically-optimized model selection, noise filtering, and window size selection using machine learning models, the method comprising:
receiving a first subset of time series data, wherein the first subset of times series data comprises values for load amounts for processing requests on computer nodes;
receiving a prediction horizon, wherein the prediction horizon comprises a future date for predicting a load amount for processing requests;
generating a feature input based on the first subset of time series data and the prediction horizon, wherein the feature input comprises a vectorized representation of the first subset of time series data and the prediction horizon;
inputting the feature input into a machine learning model, wherein the machine learning model includes:
a first model component, wherein the first model component is trained to select, for given time series data and given prediction horizons, an optimal base model collection for an ensemble of base models and select an optimal parameter set for the optimal base model collection, and wherein the optimal base model collection and optimal parameter set are selected by evaluating multiple candidate model structures and associated parameters to minimize prediction error and maximize a predictive performance metric;
a second model component, wherein the second model component is trained to select a filtering parameter for the given time series data and given prediction horizons; and
a third model component, wherein the third model component is trained to select an optimal window size for the given time series data and given prediction horizons;
receiving an output from the machine learning model; and
generating for display, on a user interface, a prediction for the first subset of time series data at the prediction horizon based on the output, wherein the prediction comprises a graphical representation of a predicted set of values for time series values and a confidence level for each of the predicted set of values for time series values.
3. The method of claim 2 , wherein the ensemble of base models comprises individual models and models based on combinations of the individual models.
4. The method of claim 2 , wherein the ensemble of base models comprises all permutations of individual models and models based on combinations of all the permutations of the individual models.
5. The method of claim 2 , wherein the second model component comprises an empirically-optimized Gaussian smoothing filter, wherein empirically-optimized the Gaussian smoothing filter comprises optimizing the filter parameter and the optimal parameter set using a single function.
6. The method of claim 2 , further comprising training the machine learning model using four-fold cross-validation.
7. The method of claim 2 , wherein variables for the first model component include values of the first subset of time series data, the prediction horizon, or the optimal window size.
8. The method of claim 2 , wherein variables for the second model component include values of the first subset of time series data, the prediction horizon, or the optimal window size.
9. The method of claim 2 , wherein variables for the third model component include values of the first subset of time series data, the prediction horizon, or the optimal parameter set.
10. The method of claim 2 , wherein the second model component is trained to filter both stationary and non-stationary noise from the given time series data.
11. The method of claim 2 , wherein the prediction includes a predicted set of values for time series values and a confidence level for each of the predicted set of values for time series values.
12. A non-transitory, computer readable medium for generating optimal data predictions in real-time for time series data signals based on empirically-optimized model selection, noise filtering, and window size selection using machine learning models comprising instructions that when executed by one or more processors, cause operations comprising:
receiving a first subset of time series data, wherein the first subset of times series data comprises values for load amounts for processing requests on computer nodes;
receiving a prediction horizon, wherein the prediction horizon comprises a future date for predicting a load amount for processing requests;
generating a feature input based on the first subset of time series data and the prediction horizon, wherein the feature input comprises a vectorized representation of the first subset of time series data and the prediction horizon;
inputting the feature input into a machine learning model, wherein the machine learning model includes:
a first model component, wherein the first model component is trained to select, for given time series data and given prediction horizons, an optimal base model collection for an ensemble of base models and select an optimal parameter set for the optimal base model collection, and wherein the optimal base model collection and optimal parameter set are selected by evaluating multiple candidate model structures and associated parameters to minimize prediction error and maximize a predictive performance metric;
a second model component, wherein the second model component is trained to select a filtering parameter for the given time series data and given prediction horizons; and
a third model component, wherein the third model component is trained to select an optimal window size for the given time series data and given prediction horizons;
receiving an output from the machine learning model; and
generating for display, on a user interface, a prediction for the first subset of time series data at the prediction horizon based on the output, wherein the prediction comprises a graphical representation of a predicted set of values for time series values and a confidence level for each of the predicted set of values for time series values.
13. The non-transitory, computer readable medium of claim 12 , wherein the ensemble of base models comprises individual models and models based on combinations of the individual models.
14. The non-transitory, computer readable medium of claim 12 , wherein the ensemble of base models comprises all permutations of individual models and models based on combinations of all the permutations of the individual models.
15. The non-transitory, computer readable medium of claim 12 , wherein the second model component comprises an empirically-optimized Gaussian smoothing filter, wherein empirically-optimized the Gaussian smoothing filter comprises optimizing the filtering parameter and the optimal parameter set using a single function.
16. The non-transitory, computer readable medium of claim 12 , further comprising training the machine learning model using four-fold cross-validation.
17. The non-transitory, computer readable medium of claim 12 , wherein variables for the first model component include values of the first subset of time series data, the prediction horizon, or the optimal window size.
18. The non-transitory, computer readable medium of claim 12 , wherein variables for the second model component include values of the first subset of time series data, the prediction horizon, or the optimal window size.
19. The non-transitory, computer readable medium of claim 12 , wherein variables for the third model component include values of the first subset of time series data, the prediction horizon, or the optimal parameter set.
20. The non-transitory, computer readable medium of claim 12 , wherein the second model component is trained to filter both stationary and non-stationary noise from the given time series data.