How to find the variables involved in the decision to report adverse action when the origination scorecard is developed using Machine Learning - XGBOOST with monotonic constraints (80 variables)

Scenario 1: easy one- Which 3 distinct noncorrelated variables are the major contributor to the rejection

Scenario 2: Little complex - Selection of variables for adverse action and then matching with development data to get two names. For an example How to get the explanation for a person rejected for credit product in below format. ?

”This person is rejected because their months since most recent inquiry with value 2, number of trade lines opened in the last 12 months with value 4, and number of satisfied trade lines with value 2 are similar to Mike Brown and Alice Wong in the datasets. Wong and Brown could not payoff.”

Disclaimer: Above Line is used from FICO website and intention is to explain my problem.

  • $\begingroup$ What is your question? If it is to "how to formulate a paragraph similar to the one you stated" then you simply make a sentence about your inputs, in this case 2-12-4-2 for different parameters (probably those most similar to Brown and Wong's) and you find the two nearest neighbours with a similar outcome and state them, i.e. "Brown and Wong". $\endgroup$ – Attack68 Jul 10 '18 at 6:47
  • $\begingroup$ @Attack68 Thanks for your comment. Question is how to achieve Adverse action in Machine Learning model. With the Logistic Regression model (limited number of Variables), it is easy to find the principal reasons for adverse action (application Rejection). In machine learning models there are many variables and complex interactions. How to find the top 5 reasons for the Rejection of Application. Many a time top variables are of a similar nature. In general how people address Adverse action while using machine learning credit score- SVM, Random Forest, XGBOOST. $\endgroup$ – June Jul 11 '18 at 2:12

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.