Lift and Gain Chart
Lift Chart and Gain Chart · Also known as: Cumulative Gain Chart, Lift Curve
Lift and gain charts visualize classifier performance by showing how much better the model performs compared to random selection, particularly useful for ranking or scoring tasks where you select a top percentage of samples. They are widely used in marketing, credit scoring, and fraud detection.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use lift and gain charts for ranking problems and cost-conscious applications where you select a limited number of samples. They are essential in marketing (how many responders are in the top 10% of targeted customers?), fraud detection (what fraction of fraud is in the top 20% of suspicious accounts?), and credit scoring (how many defaults are in the riskiest 15%?).
Strengths & limitations
- Directly shows practical value: how many positives you capture by selecting top-ranked samples.
- Useful for cost-sensitive decisions where examining all samples is infeasible.
- Intuitive visualization making results easy to communicate to non-technical stakeholders.
- Works well on imbalanced datasets where baseline performance is low.
- Does not provide a single numeric score for model comparison (though lift values can be extracted).
- Sensitive to sample size and class imbalance; lift values vary with the proportion of positives.
- Requires probability outputs or confidence scores, not just class predictions.
- Can be misleading if interpreted without examining the underlying model performance.
Frequently asked
What does a lift of 2.5 mean?
A lift of 2.5 at a given percentile means the model is 2.5 times better than random selection at identifying positives within that percentage of the ranked population. For example, lift 2.5 at 20% means if you examine the top 20% of ranked samples, you will find 2.5 times as many positives as you would by randomly selecting 20%.
How do lift and gain charts differ?
Lift charts show the ratio of model performance to random performance at each percentile. Gain charts show the cumulative percentage of positives found as you move through the ranked list. Both convey similar information but lift is more directly interpretable as 'relative improvement,' while gain is more intuitive as 'how many have I found so far.'
Can I use lift and gain charts for multi-class problems?
Yes, but the interpretation changes. For multi-class, you typically evaluate each class separately: rank by the probability of a specific class being positive, then compute lift and gain for that class. This produces one chart per class of interest.
Sources
- Maimon, O. Z., & Rokach, L. (Eds.). (2010). Data Mining and Knowledge Discovery Handbook (2nd ed.). Springer. DOI: 10.1007/978-0-387-09823-4 ↗
- Naeem Siddiqi (2006). Credit Risk Scorecards: Developing and Implementing Intelligent Credit Scoring. John Wiley & Sons. link ↗
How to cite this page
ScholarGate. (2026, June 3). Lift Chart and Gain Chart. ScholarGate. https://scholargate.app/en/model-evaluation/lift-and-gain-chart
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Precision-Recall AUCModel Evaluation↔ compare
- Recall (Sensitivity)Model Evaluation↔ compare