Back to results
US 2026/0300761 A1Published application
Generating gradient boosting machine model reason codes
AssigneeCapital One Financial Corporation
InventorsBingrong Jiang, Taisong Zou
Life of this patent
Estimating the term…
Abstract
Techniques for providing a prioritization order of a set of features of a prediction model involve (i) determining feature contribution values for a plurality of possible outcomes of the prediction model, (ii) storing the determined feature contribution values, (iii) retrieving feature contribution values that are associated with a given outcome that is predicted by the prediction model, and (iv) utilizing the retrieved feature contribution values to determine the respective contributions of the features to the given outcome.
Claims20
A computing platform comprising:
- at least one processor;
- at least one non-transitory computer-readable medium;
- and
- program instructions stored on the at least one non-transitory computer-readable medium that, when executed by the at least one processor, cause the computing platform to:
- for a prediction model comprising multiple sub-models that are each configured to predict a sub-model outcome from input data for a set of features:
- determine, for each respective sub-model of the multiple sub-models, respective per-outcome feature contribution information for a plurality of possible sub-model outcomes of the respective sub-model;
- and
- store, for each respective sub-model of the multiple sub-models, the respective per-outcome feature contribution information for the plurality of possible sub-model outcomes of the respective sub-model;
- utilize the prediction model to predict a given outcome from given input data for the set of features, wherein the given outcome is based on a given combination of sub-model outcomes predicted by the multiple sub-models;
- retrieve, from the stored per-outcome feature contribution information, feature contribution information associated with the sub-model outcomes in the given combination of sub-model outcomes predicted by the multiple sub-models;
- and
- based on the retrieved feature contribution information, determine respective contributions of the set of features to the given outcome.
9 more elements
A computing platform comprising:
- at least one processor;
- at least one non-transitory computer-readable medium;
- and
- program instructions stored on the at least one non-transitory computer-readable medium that, when executed by the at least one processor, cause the computing platform to:
- for a prediction model that is configured to predict an outcome from input data for a set of features:
- determine per-outcome feature contribution information for a plurality of possible outcomes associated with the prediction model;
- and
- store the per-outcome feature contribution information for the plurality of possible outcomes;
- utilize the prediction model to predict a given outcome from given input data for the set of features;
- retrieve, from the stored per-outcome feature contribution information, feature contribution information associated with the given outcome;
- and
- based on the retrieved feature contribution information, determine respective contributions of the set of features to the given outcome.
9 more elements
A computer-implemented method comprising:
- for a prediction model comprising multiple sub-models that are each configured to predict a sub-model outcome from input data for a set of features:
- determining, for each respective sub-model of the multiple sub-models, respective per-outcome feature contribution information for a plurality of possible sub-model outcomes of the respective sub-model;
- and
- storing, for each respective sub-model of the multiple sub-models, the respective per-outcome feature contribution information for the plurality of possible sub-model outcomes of the respective sub-model;
- utilizing the prediction model to predict a given outcome from given input data for the set of features, wherein the given outcome is based on a given combination of sub-model outcomes predicted by the multiple sub-models;
- retrieving, from the stored per-outcome feature contribution information, feature contribution information associated with the sub-model outcomes in the given combination of sub-model outcomes predicted by the multiple sub-models;
- and
- based on the retrieved feature contribution information, determining respective contributions of the set of features to the given outcome.
5 more elements
Dependencies are inferred from the claim text.