¶ Description
Create a Kolmogorov-Smirnov curve after applying a model.
¶ Parameters
¶ Parameters tab
Parameters:
- Chart Title
- Decile Column
- Proportion of "Bad" (1)
- Proportion of "Good" (0)
- Point Size
- Color for "Bad"
- Color for "Good"
- Save Image in PNG? (empty field=no save)
- Plot Size (higher = smaller fonts)
¶ Description tab
Parameters:
- Script name
- Short description
- Revision
- Description
¶ Configuration tab
See dedicated page for more information.
¶ About
Create a Kolmogorov-Smirnov curve after applying a model. The input data consists of an aggregated table (usually deciles) in which we have computed:
- Proportion of “bad” in each decile (usually the target)
- Proportion of “good” in each decile
- The decile
This curve is often used in credit risk analysis, but is – generally speaking – not a very robust way to evaluate a model as it assumes a normal distribution of the scores which is not appropriate if you work with actual transactional data. Some authors suggest that a logartimic transformation can improve it, but this would only be appropriate if using a logistic regression (which depends on transformed scores, and in general WoE), but not with any ML algortithm that do not use this transformation.
¶ Use Cases
- Evaluate scorecard model effectiveness
- Detect class separation across population bins
- Visualize classification accuracy for binary labels
