Research map / Graph learning / Graph algorithms and clustering
Uplift Model Evaluation with Ordinal Dominance Graphs
Brecht Verbeken, Marie-Anne Guerry, Wouter Verbeke, Sam Verboven
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace Qini score at 10% with pROCini score | uplift evaluation metric · Hillstrom semisynthetic, S_LR > T_LR | % runs ranked correctly: 50.4570 -> 62.5246 | Improved |
| replace Qini score with pROCini score | uplift evaluation metric · Hillstrom semisynthetic, S_LR > T_LR | % runs ranked correctly: 58.8713 -> 62.5246 | Improved |
| replace Qini score with pROCini score | uplift evaluation metric · Hillstrom semisynthetic, T_LR > S_XGB | % runs ranked correctly: 54.2210 -> 55.4111 | Improved |
| replace Qini score with pROCini score | uplift evaluation metric · Hillstrom semisynthetic, S_XGB > T_XGB | % runs ranked correctly: 55.4796 -> 57.6352 | Improved |
| replace Qini score with pROCini score | uplift evaluation metric · Hillstrom semisynthetic, true_ITE > S_LR | % runs ranked correctly: 50.9785 -> 51.4539 | Improved |
| use ROCini or pROCini instead of the ROC-curve-free Qini in the simulation | uplift evaluation metric · Synthetic beta-distributed outcomes | % runs ranking perfect model first | Improved |
| use pROCini score when control-group positive probability is low, (alpha,beta)=(5,25) | pROCiniS/ROCiniS/CROCS · Synthetic simulation, (alpha,beta)=(5,25) | % runs ranking perfect model first | Hurt |
| increase population size N from 1000 to 16000 | Qini score distribution · Simulated Qini score | variance: 78 × 10−6 -> 5 × 10−6 | Improved |
| increase population size N from 1000 to 16000 | Qini score distribution · Simulated Qini score | mean: 0.711 -> 0.724 | No clear effect |
| increase signal strength v | Qini score · Simulated Qini score | Qini score | Improved |
| add larger model error E | Qini score · Simulated Qini score | Qini score | Hurt |
| use S-learner instead of T-learner with logistic regression | uplift model · Hillstrom semisynthetic ITE recovery | MSE vs true ITE: 0.0098 -> 0.0030 | Improved |
Closest papers by meaning
- Rethinking Causal Ranking: A Balanced Perspective on Uplift Model Evaluation (ICML 2025 · similarity 0.89)
- Imbalance-Aware Uplift Modeling for Observational Data (AAAI 2022 · similarity 0.83)
- Invariant Deep Uplift Modeling for Incentive Assignment in Online Marketing via Probability of Necessity and Sufficiency (ICML 2025 · similarity 0.82)
- SLACE: A Monotone and Balance-Sensitive Loss Function for Ordinal Regression (AAAI 2025 · similarity 0.80)
- Identifiability Matters: Revealing the Hidden Recoverable Condition in Unbiased Learning to Rank (ICML 2024 · similarity 0.80)
- Causal Discovery via Bayesian Optimization (ICLR 2025 · similarity 0.79)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
