Skip to content
ML Atlas

Credit Scoring

Predicting how likely a borrower is to repay, traditionally with interpretable scorecards based on logistic regression, increasingly with gradient boosting, under strict fairness and explainability rules.

Also credit score · FICO score · scorecard · loan default prediction · credit risk model

Fig. / How it works
Loan applicationLoan applicationHistory, debt, incomeHistory, debt, incomeScorecard modelScorecard modelCredit scoreCredit score
Role · kind
Real systemsFamous system
Level
Beginner

Fair, Isaac and Company scorecards from 1958; general-purpose FICO score in 1989

A bank looks at how you have handled money before: paying bills on time, how much you owe. It turns that into a number that guesses how likely you are to pay back a new loan.

Where it fits

Classical Machine LearningClassical Machine LearningExplainabilityExplainabilityGradient BoostingGradient BoostingLogistic RegressionLogistic RegressionCredit Scoring
+All relationships as a list

Is a kind of / part of

See it on the full map

How it works

Lenders have used statistical credit scoring for decades. Fair, Isaac and Company (now FICO) built custom scoring systems from 1958 and introduced a general-purpose credit bureau score in 1989, the familiar three-digit number.

Classic scorecards are built with logistic regression:

  • Each factor (payment history, how much credit you use, length of history) is split into ranges.
  • Each range earns points; the points add up to the score.

Because the model is additive, anyone can see why a score is low, which regulators require: in many countries a lender must tell you the main reasons for a rejection.

Modern models often use gradient boosting, which is more accurate but less transparent. Lenders pair it with explainability methods like SHAP to produce reasons for each decision.

The fairness stakes are high. Even without using protected attributes, a model can discriminate through proxies such as postcode (bias and fairness). That is why the EU AI Act lists creditworthiness assessment of individuals as a high-risk use with strict requirements.

Everyday use cases

  • Case 01

    How you've probably met it

    When applying for a credit card, mortgage, phone contract or "buy now, pay later" at checkout.

Strengths, weaknesses and when to use it

Strengths

  • Faster, more consistent lending decisions than manual review.

Weaknesses

  • Can encode historical discrimination through proxy features; people with little credit history are poorly served.

When to use itAs the everyday example of a high-stakes, regulated classifier where explanations are mandatory.

Compare with

Go deeper

Sources

  1. [1]Wikipedia: Credit score in the United States
  2. [2]Regulation (EU) 2024/1689 (AI Act), Annex III

Facts checked as of 2026-10-12