Uncertainty and models
Lift chart
Shows outcome concentration in selected cases relative to a baseline.
- Chart
- Model evaluation
- Ranking
What it answers
How much better is selection than the baseline rate?
- Shape
- Lift across score-ranked groups
- What it shows
- Outcome enrichment in ranked selections.
Example
The question
How much better than random is our model at finding buyers?
Response rate by score decile, relative to average (lift chart)
Illustrative data. It belongs to no company.
When to use it
Use it when
Evaluating targeting or prioritisation rankings.
Avoid it when
Undefined baseline or causal claims about model treatment effects.
How to read it
Read lift relative to baseline 1 and check cumulative/group-specific definition.
Note
Define cumulative versus group-specific lift.
The data you need
- Layout
- Long table
- Fields
- CaseID: text; Actual: 0/1; Score: number
- One row is
- One case with its true binary outcome and prediction score. What is the grain?
- Rules
- Define the positive class and score direction. Use held-out predictions for evaluation. Calibration needs probabilities in [0,1].
- Optional
- Weight. ROC/PR alternatives: precomputed Threshold,TP,FP,TN,FN with complete threshold coverage.
- Preparation
- Rank cases by score, define baseline outcome rate and calculate group-specific or cumulative lift.
- Units
- Lift =selected-group outcome rate/overall baseline outcome rate. Lift 1 means baseline performance.
Worked example
Overall 10 positives in 100 cases gives 10% baseline. Top 20 contain 6 positives:30% rate and 3 x lift.
| CaseID | Actual | Score |
|---|---|---|
| A | 1 | 0.90 |
| B | 0 | 0.20 |
| C | 1 | 0.65 |
| D | 0 | 0.55 |