Systems and methods for business analytics management and modeling
AssigneePrevedere Inc.
InventorsKyler Cooper, Jessica Emily Dolezal, Andrew Duguay, Alexander C. Elek, Danielle Marceau, Richard Wagner
Life of this patent
Estimating the term…
Abstract
The present invention relates to systems and methods for model generation. The model is generated by selecting indicators that are relevant to the model, determining a strength score for each of the indicators, ranking the indicators by their strength scores, and bucketizing the indicators. Different permutations of the indicators are then selected for modeling in parallel. The model results are compared, and the ‘best’ model (most historically accurate) is selected for display within a report.
Claims18
A computerized method for generating a model, useful in association with a business analytics system, the method comprising:
- selecting a plurality of indicators relevant to a model;
- generating a strength score for each of the plurality of indicators using a processor that calculates a R-squared value and procyclic values for each indicator to derive the strength scores by adding the R-squared value and procyclic values and multiplying the results by a seasonality determination, a normalized fraction of a number of model indicators over a total number of models, a normalized difference in a last update, a normalized difference between a minor outlier count and a major outlier count, and a fraction of data overlap over a frequency value of the indicator;
- ranking the plurality of indicators by their strength score;
- bucketizing the plurality of ranked indicators;
- configuring a plurality of models using machine learning;
- modeling various permutations of the bucketized indicators in parallel to generate a plurality of model forecasts;
- selecting one of the plurality of model forecasts with the highest accuracy;
- and
- generating a report using the selected one model forecast.
A computerized system for generating a model comprising:
- a data aggregation server for selecting a plurality of indicators relevant to a model;
- a data analyzer for generating a strength score for each of the plurality of indicators using a processor that calculates a R-squared value and procyclic values for each indicator to derive the strength scores by adding the R-squared value and procyclic values and multiplying the results by a seasonality determination, a normalized fraction of a number of model indicators over a total number of models, a normalized difference in a last update, a normalized difference between a minor outlier count and a major outlier count, and a fraction of data overlap over a frequency value of the indicator;
- a modeling engine for configuring a plurality of models using machine learning, modeling various permutations of the bucketized indicators in parallel to generate a plurality of model forecasts, and selecting one of the plurality of model forecasts with the highest accuracy;
- and
- a reporting module for generating a report using the selected one model forecast.
Dependencies are inferred from the claim text.