Systems and methods for risk factor predictive modeling
A system and method for Medical Claims Risk Score (MCRS) algorithmic underwriting includes a predictive machine learning model configured to generate underwriting decisions on electronic applications. MCRS underwriting applies word embedding modeling, such as GloVe (global vectors), to transform high dimensional MC records into single-code word vectors. These single-code word vectors are employed in regression modeling, and may include summarized embedding coordinates aggregated at the applicant level. Regression modeling uses medical claim codes data and underwriting decision data stored for historical underwriting applicants to train a random forest model to predict relative mortality risk for underwriting applicants. A risk rating may be derived from the underwriting decision data based upon standard quantitative risk ratings of a plurality of risk classes. Other inputs to the random forest model may include cohort level applicant profile data, such as applicant issue age and sex.
1 . A computer-implemented method comprising:
training, by a server, a predictive machine learning model to receive medical claims codes data identified with applicants of the electronic applications and determine a score for at least one electronic applications that classifies the at least one electronic application into one of a plurality of groups based upon the score,
wherein the predictive machine learning model is trained by inputting, for each applicant record of a plurality of applicant records associated with respective historical underwriting applications, medical claims codes data for the respective applicant record into a word embedding model, wherein the word embedding model:
encodes relationships amongst vector offsets representing the medical claims codes data;
reduces a dimensionality of the relationships amongst vector offsets by:
generating, based on the encoded relationships, embedding coordinates in two or three dimensions for the vector offsets representing the medical claims codes data for each respective applicant record of the plurality of applicant records; and
selecting a minimum value and a maximum value for each of the dimensions of the embedding coordinates for the medical claims codes data for the respective applicant record; and
responsive to the minimum values and the maximum values,
upon receiving a new electronic application from a user device, retrieving, by the server, medical claims code data identified with an applicant of the new electronic application;
executing, by the server, the trained predictive machine learning model using the electronic application:
classifying, by the server using the predictive machine learning model, the electronic application;
and
generating, by the server, an explanation based on the medical claims codes associated with the minimum values and the maximum values that resulted in a classification of the electronic application.
2 . The method of claim 1 , wherein the word embedding model comprises a Global Vectors (GloVe) model.
3 . The method of claim 1 , wherein the predictive machine learning model further comprises a regression model in combination with the word embedding model, wherein the regression model is trained by inputting underwriting decisions data associated with the plurality of applicant records.
4 . The method of claim 3 , wherein the regression model is trained to determine the score by inputting the underwriting decisions data associated with the plurality of applicant records and vector offsets generated by the word embedding model for the medical claims codes data for the plurality of applicant records.
5 . The method of claim 3 , wherein the underwriting decisions data for each of the plurality of applicant records comprises a score corresponding to one of the plurality of groups.
6 . The method of claim 3 , wherein the regression model comprises a random forests ensemble learning method.
7 . The method of claim 1 , wherein the embedding coordinates have a number of dimensions in a range of 35 to 200.
8 . The method of claim 1 , wherein the predictive machine learning model produces a SHAP contribution for every minimum value and maximum value, and generating an explanation is further based on the SHAP contributions.
9 . The method of claim 1 , wherein the predictive machine learning model is further configured to receive demographic data and the electronic application is classified responsive to the minimum values, the maximum values, and the demographic data.
10 . A computer-implemented method comprising:
training by a server, a predictive machine learning model to receive medical claims codes data identified with applicants of the electronic applications and determine a score for at least one electronic applications that classifies the at least one electronic application into one of a plurality of groups based upon the score,
wherein the predictive machine learning model is trained by inputting, for each applicant record of a plurality of applicant records associated with respective historical underwriting applications, medical claims codes data and underwriting decision data for the respective applicant record into a word embedding model in combination with a regression model, wherein the word embedding model:
encodes relationships amongst vector offsets representing the medical claims codes data;
reduces a dimensionality of the relationships amongst vector offsets by:
generating, based on the encoded relationships, embedding coordinates in two or three dimensions for the vector offsets representing the medical claims codes data for each respective applicant record of the plurality of applicant records; and
selecting a minimum value and a maximum value for each of the dimensions of the embedding coordinates for the medical claims codes data for the respective applicant record; and
responsive to the minimum values and the maximum values,
classifying, by the server using the predictive machine learning model, the electronic application;
upon receiving a new electronic application from a user device, retrieving, by the server, medical claims code data identified with an applicant of the new electronic application:
executing, by the server, the trained predictive machine learning model using the electronic application:
generating, by the server, an explanation based on the medical claims codes associated with the minimum values and the maximum values that resulted in a classification of the electronic application; and
presenting, by the server, a user interface for display on the user device an offer based upon the one of the plurality of groups and the explanation.
11 . The method of claim 10 , wherein the regression model is trained by inputting the underwriting decision data and vector offsets generated by the trained word embedding model for the medical claims codes data for the plurality of applicant records.
12 . The method of claim 10 , wherein the predictive machine learning model produces a SHAP contribution for every minimum value and maximum value, and generating an explanation is further based on the SHAP contributions.
13 . A system, comprising:
a non-transitory machine-readable memory configured to store medical claims codes data for a plurality of applicant records associated with respective historical underwriting applications;
a predictive machine learning model configured to receive the medical claims codes data identified with an applicant of an electronic application and determine a score for the electronic application that classifies the electronic application into one of a plurality of groups based upon the score, wherein the predictive machine learning model is trained for each applicant record of the plurality of applicant records by inputting medical claims codes data for the respective applicant record into a word embedding model; and
a processor in communication with the non-transitory, machine-readable memory and the predictive modeling model, configured to execute a set of instructions instructing the processor to:
upon receiving an electronic application from a user device, retrieve medical claims code data identified with an applicant of the electronic application from the non-transitory machine-readable memory;
determine the score for the electronic application that classifies the electronic application into one of the plurality of groups based upon the score, wherein determining the score comprises:
encoding relationships amongst vector offsets representing the medical claims codes data;
reduces a dimensionality of the relationships amongst vector offsets by:
generating, based on the encoded relationships, embedding coordinates in two or three dimensions for the vector offsets representing the medical claims codes data for each respective applicant record of the plurality of applicant records;
selecting a minimum value and a maximum value for each of the dimensions of the embedding coordinates for the medical claims codes data for the respective applicant record; and
determining the score for the electronic application based on the maximum values and the minimum values; and
responsive to determining the score,
classify the electronic application into one of the plurality of groups based on the score;
train a predictive machine learning model to receive medical claims codes data identified with applicants of electronic applications and determine a score for at least one electronic applications that classifies the at least one electronic application into one of a plurality of groups based upon the score;
generate an explanation for classifying the electronic application into the group based on the medical claims codes associated with the minimum values and the maximum values that resulted in a classification of the electronic application; and
present a user interface for display on the user device an offer based upon the one of the plurality of groups and the explanation.
14 . The system of claim 13 , wherein the word embedding model comprises a Global Vectors (GloVe) model.
15 . The system of claim 13 , further comprising a regression model, wherein the non-transitory machine-readable memory further stores underwriting decision data for the plurality of applicant records, wherein the regression model is trained by inputting vector offsets generated by the word embedding model for the medical claims codes data for the plurality of applicant record, and by inputting the underwriting decisions data associated with the plurality of applicant records.
16 . The system of claim 15 , wherein the underwriting decision data for each of the plurality of applicant records comprises a score corresponding to one of the plurality of groups.
17 . The system of claim 15 , wherein the regression model comprises a random forests ensemble learning method.
18 . The system of claim 13 , wherein the embedding coordinates have a number of dimensions in a range of 35 to 200.
19 . The system of claim 13 , wherein the processor produces a SHAP contribution for every minimum value and maximum value, and generating an explanation is further based on the SHAP contributions.
20 . The system of claim 13 , wherein the processor is further configured to retrieve demographic information identified with the applicant of the electronic application and the electronic application is further classified based on the demographic information.