IP Library › Granted Patent US 12664804

Granted Patent B2

US 12664804 · App. 18/351,792 · Granted Jun 23, 2026

Predictive machine learning feature selection

Inventors: Karim M Mahmoud Mohamed Moustafa (Dublin, IE); Eugene Edward Farrell (Galway, IE); Lisa E. Walsh (Dublin, IE); Harutyun Shahumyan (Dublin, IE); Smitashree Choudhury (Dublin, IE); Arjit Agrawal (Dublin, IE)

Assignee: Optum Services (Ireland) Limited

G06V30/19013 G06V30/19167

View Patent ↗

Loading inventors, assignments & file history…

Monitor This Case

Get email alerts when status or documents change.

Order Certified Copies

Most orders are placed with the USPTO same day — all within 24 business hours.

Order via The Patent Place →

Pre-filled with this patent's details

Quick Facts

Patent No.: US 12664804
App. No.: 18/351,792
Granted: Jun 23, 2026
Kind: B2

External Links

USPTO Patent Center↗ Full Patent PDF↗

Abstract

Various embodiments of the present disclosure provide feature engineering techniques for improving machine learning model development, performance, and maintenance. The feature engineering techniques include generating a model description vector from a textual model description for a target machine learning model and using the model description vector to generate one or more description-based similarity vectors that each include one or more of a plurality of similarity scores for a plurality of machine learning features. The feature engineering techniques include generating a label-based similarity vector based on a comparison between training data for the target machine learning model and a plurality of feature values of the plurality of machine learning features. The feature engineering techniques include providing a predictive feature set for the target machine learning model based on the one or more similarity vectors and the one or more label-based similarity vectors.

Claims (71)

1 . A computer-implemented method comprising:

generating, by one or more processors, a model description vector from a textual model description for a target machine learning model;

generating, by the one or more processors, a plurality of description-based similarity vectors based on the model description vector, wherein:

(i) a description-based similarity vector of the plurality of description-based similarity vectors comprises one or more of a plurality of similarity scores for a plurality of machine learning features, and

(ii) the plurality of similarity scores is based on the model description vector;

generating, by the one or more processors, a label-based similarity vector based on a comparison between training data for the target machine learning model and a plurality of feature values of the plurality of machine learning features;

generating, by the one or more processors and based on the plurality of description-based similarity vectors and the label-based similarity vector, a predictive feature set comprising a pre-defined number of features that provide improved model performance when training the target machine learning model; and

training, by the one or more processors, the target machine learning model using the predictive feature set.

2 . The computer-implemented method of claim 1 , wherein the description-based similarity vector comprises a problem-based similarity vector and the plurality of similarity scores is based on a comparison between the model description vector and a plurality of feature description vectors corresponding to a plurality of textual feature descriptions for the plurality of machine learning features.

3 . The computer-implemented method of claim 2 further comprising:

generating, using a language processing model, the plurality of feature description vectors based on the plurality of textual feature descriptions for the plurality of machine learning features;

generating, using a vector similarity model, a plurality of problem-based similarity scores based on a comparison between the model description vector and the plurality of feature description vectors; and

generating the problem-based similarity vector based on the plurality of problem-based similarity scores, wherein the problem-based similarity vector comprises one or more similarity scores of the plurality of problem-based similarity scores that satisfy a relative similarity threshold.

4 . The computer-implemented method of claim 3 , wherein the model description vector is generated using the language processing model.

5 . The computer-implemented method of claim 1 , wherein the description-based similarity vector comprises a domain-based similarity vector, and generating the plurality of description-based similarity vectors comprises:

identifying one or more relevant natural language documents based on a plurality of document similarity scores, wherein the plurality of document similarity scores is based on a comparison between the model description vector and a plurality of document vectors corresponding to a plurality of natural language documents for a model domain;

generating, using a vector similarity model, a plurality of domain-based similarity scores based on a comparison between (i) one or more document vectors of the plurality of document vectors that correspond to the one or more relevant natural language documents and (ii) a plurality of feature description vectors corresponding to a plurality of textual feature descriptions for the plurality of machine learning features; and

generating the domain-based similarity vector based on the plurality of domain-based similarity scores, wherein the domain-based similarity vector comprises one or more domain-based similarity scores of the plurality of domain-based similarity scores that satisfy a relative similarity threshold.

6 . The computer-implemented method of claim 1 , wherein the description-based similarity vector comprises a historical model-based similarity vector, and the plurality of description-based similarity vectors comprises:

identifying one or more relevant models based on a plurality of model similarity scores, wherein the plurality of model similarity scores is based on a comparison between the model description vector and a plurality of model vectors corresponding to a plurality of historical model descriptions for a plurality of historical machine learning models; and

generating the historical model-based similarity vector based on the one or more relevant models and the plurality of model similarity scores.

7 . The computer-implemented method of claim 1 , wherein the training data comprises one or more training inputs and one or more training outputs, and wherein generating the label-based similarity vector comprises:

identifying the plurality of feature values based on the one or more training inputs;

generating a plurality of feature association metrics for the plurality of machine learning features based on a comparison between the plurality of feature values and the one or more training outputs; and

generating the label-based similarity vector based on the plurality of feature association metrics, wherein the label-based similarity vector comprises one or more of the plurality of feature association metrics that satisfy a relative similarity threshold.

8 . The computer-implemented method of claim 1 , wherein the plurality of description-based similarity vectors comprises a domain-based similarity vector, a problem-based similarity vector, and a historical model-based similarity vector, and wherein the computer-implemented method further comprises:

generating a plurality of aggregated feature relevance scores for a subset of machine learning features from the plurality of machine learning features based on an aggregation of the domain-based similarity vector, the problem-based similarity vector, the historical model-based similarity vector, and the label-based similarity vector; and

generating the predictive feature set based on the plurality of aggregated feature relevance scores.

9 . The computer-implemented method of claim 8 , wherein the domain-based similarity vector, the problem-based similarity vector, the historical model-based similarity vector, and the label-based similarity vector are associated with one or more different machine learning features from the plurality of machine learning features.

10 . The computer-implemented method of claim 9 , wherein the subset of machine learning features from the plurality of machine learning features comprises one or more machine learning features from each of the domain-based similarity vector, the problem-based similarity vector, the historical model-based similarity vector, and the label-based similarity vector.

11 . The computer-implemented method of claim 9 further comprising:

generating a plurality of scaled feature relevance scores by applying one or more scaling parameters to the plurality of aggregated feature relevance scores, wherein the one or more scaling parameters include one or more user defined parameters indicative of a user preference for each of the domain-based similarity vector, the problem-based similarity vector, the historical model-based similarity vector, and the label-based similarity vector; and

generating the predictive feature set based on the plurality of scaled feature relevance scores.

12 . A system comprising:

one or more processors; and

at least one memory storing processor-executable instructions that, when executed by the one or more processors, cause the one or more processors to:

generate a model description vector from a textual model description for a target machine learning model;

generate a plurality of description-based similarity vectors based on the model description vector, wherein:

(ii) the plurality of similarity scores is based on the model description vector;

generate a label-based similarity vector based on a comparison between training data for the target machine learning model and a plurality of feature values of the plurality of machine learning features;

generate, based on the plurality of description-based similarity vectors and the label-based similarity vector, a predictive feature set comprising a pre-defined number of features that provide improved model performance when training the target machine learning model; and

train the target machine learning model using the predictive feature set.

13 . The system of claim 12 , wherein the description-based similarity vector comprises a problem-based similarity vector and the plurality of similarity scores is based on a comparison between the model description vector and a plurality of feature description vectors corresponding to a plurality of textual feature descriptions for the plurality of machine learning features.

14 . The system of claim 13 , wherein the one or more processors are further configured to:

generate, using a language processing model, the plurality of feature description vectors based on the plurality of textual feature descriptions for the plurality of machine learning features;

generate, using a vector similarity model, a plurality of problem-based similarity scores based on a comparison between the model description vector and the plurality of feature description vectors; and

generate the problem-based similarity vector based on the plurality of problem-based similarity scores, wherein the problem-based similarity vector comprises one or more similarity scores of the plurality of problem-based similarity scores that satisfy a relative similarity threshold.

15 . The system of claim 12 , wherein the description-based similarity vector comprises a domain-based similarity vector, and generating the plurality of description-based similarity vectors comprises:

16 . The system of claim 12 , wherein the description-based similarity vector comprises a historical model-based similarity vector, and the plurality of description-based similarity vectors comprises:

generating the historical model-based similarity vector based on the one or more relevant models and the plurality of model similarity scores.

17 . One or more non-transitory computer-readable storage media including instructions that, when executed by one or more processors, cause the one or more processors to:

generate a model description vector from a textual model description for a target machine learning model;

generate a plurality of description-based similarity vectors based on the model description vector, wherein:

(ii) the plurality of similarity scores is based on the model description vector;

train the target machine learning model using the predictive feature set.

18 . The one or more non-transitory computer-readable storage media of claim 17 , wherein the training data comprise one or more training inputs and one or more training outputs, and wherein generating the label-based similarity vector comprises:

identifying the plurality of feature values based on the one or more training inputs;

19 . The one or more non-transitory computer-readable storage media of claim 17 , wherein the plurality of description-based similarity vectors comprise a domain-based similarity vector, a problem-based similarity vector, and a historical model-based similarity vector, and wherein the instructions further cause the one or more processors to:

generate a plurality of aggregated feature relevance scores for a subset of machine learning features from the plurality of machine learning features based on an aggregation of the domain-based similarity vector, the problem-based similarity vector, the historical model-based similarity vector, and the label-based similarity vector; and

generate the predictive feature set based on the plurality of aggregated feature relevance scores.

20 . The one or more non-transitory computer-readable storage media of claim 19 , wherein the domain-based similarity vector, the problem-based similarity vector, the historical model-based similarity vector, and the label-based similarity vector are associated with one or more different machine learning features from the plurality of machine learning features.