Reading GiniMachine scoring results

GiniMachine results appear on the application’s Scoring tab and again inside the Make a decision task for applications awaiting a credit decision.

Pic 4

The application record carries the following tabs:

  • General info
  • Borrower details
  • Bank info
  • Attachments
  • Contract
  • Credit bureau
  • Scoring
  • Collateral info
  • Audit trail
  • Profitability info

The Result line

The Result at the top of the Scoring tab is the output of the Scoring result decision table — the combined verdict, not GiniMachine’s. It is displayed with a color indicator.

The GiniMachine scoring block

Field Example Description
Model lb_person_scoring_9_columns The model that produced this result. Recorded on the application, so a result can be traced back to the model that generated it.
Probability of repayment 0.8090 The model’s numeric output for this application.
Resolution Approve The model’s verdict and the value passed into the Scoring result decision table.

The attribute effect table

Below the Probability of repayment, GiniMachine lists every attribute the model used, with the value taken from the application and that attribute’s effect.

Attribute Value Effect
Owns vehicle False 0.0124
Current age 36.0 0.0077
Marital status Married 0.0051
Industry sector Chemical industries 0.0042
Education Masters degree 0.0040
Length of employment > 4 years 0.0029
Owns property False 0.0023
Position Senior level manager −0.0140
Monthly income 5000.0 −0.0810

How to read it:

  • The table is sorted by effect, descending — the strongest positive contributors first, the strongest negative contributors last.
  • The sign shows the direction of each attribute’s contribution to the outcome; the magnitude shows how much weight it carried.
  • Magnitude matters more than sign when reviewing an application. In the example above, Monthly income = 5000.0 at −0.0810 carries roughly six times the weight of any single positive contributor: the application scored well overall, but income was the one attribute working against it.
  • The attributes listed are exactly the ones the model was trained on. A model named …_9_columns shows nine rows.

The attributes with the largest negative effects are the starting point for explaining a declined application to the borrower, where the market requires it.

Search documentation