QuestionQ81

VerifiedTesting AI-Specific Quality Characteristics

A bank intends to use an algorithm to decide which applicants should receive a loan. It hires a data scientist to build a logistic-regression model that predicts whether an applicant will repay the loan. The bank has sufficient historical-customer data to randomly divide it into a training data set and a test/validation data set. A logistic-regression model is built on the training data set using these independent variables:

  • Gender
  • Marital status
  • Number of dependents
  • Education
  • Income
  • Loan amount
  • Loan term
  • Credit score

The model shows that applicants with higher credit scores and greater total incomes are more likely to repay their loans. Based on previous models developed for other banks, the data scientist has suggested that the model may contain bias.

Given this information, what is the best testing approach for checking the model for potential bias?

Explanation

Bias in a lending model can be inherited from historical training data through skewed representation, inaccurate values, or relationships involving protected attributes and proxies. Exploratory data analysis of the training data can audit distributions and data quality for such sources of bias; responsible ML guidance specifically recommends auditing and exploring data for potential bias before training and evaluating outcomes for fairness.

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!