Research map / Learning theory and algorithms / Prediction and decision losses
From Predictive Methods to Missing Data Imputation: An Optimization Approach
Dimitris Bertsimas, Colin Pawlowski, Ying Daisy Zhuo
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 |
|---|---|---|---|
| cross-validate to pick among K-NN, SVM and tree objectives (opt.cv) instead of benchmark.cv | opt.impute · 84 UCI data sets, 30% MCAR | MAE: 10.1% average improvement in MAE down to 0.118 | Improved |
| use opt.knn instead of plain knn imputation | K-NN imputation · UCI imputation (30% MCAR) | delta MAE: -0.0051 (0.4574) | No clear effect |
| use opt.svm instead of plain knn imputation | K-NN imputation · UCI imputation (30% MCAR) | delta MAE: +0.0013 (0.8485) | No clear effect |
| use opt.tree instead of mean impute | imputation · UCI imputation (30% MCAR) | delta MAE: -0.0454 (<0.001***) | Improved |
| add cross-validated selection of the model on top of individual opt.impute methods (ablation) | opt.impute · UCI imputation (30% MCAR) | delta MAE vs knn: -0.0134 (0.0499*) vs -0.0051 (0.4574) for opt.knn | Improved |
| use opt.cv single imputation instead of best benchmark.cv before downstream training | downstream classification · 50% missing | out-of-sample accuracy: 84.4% -> 86.1% | Improved |
| use opt.cv single imputation instead of best benchmark.cv before downstream training | downstream regression · 50% missing | out-of-sample R2: 0.315 -> 0.339 | Improved |
| use opt.mi multiple imputation instead of mice | downstream classification · 10% to 50% missing | accuracy: -0.0001 (1.0000) at 10%; 0.0052 (<0.001***) at 50% | Mixed |
| use opt.mi multiple imputation instead of mice | downstream regression · 10% to 50% missing | R2: 0.0014 -> 0.0097 delta R2 from 10% to 50% missing | Improved |
| use multiple imputation opt.mi instead of single imputation opt.cv (ablation) | opt.impute downstream · 50% missing | accuracy and R2: 0.0038 accuracy; 0.0174 R2 | Improved |
| use block coordinate descent instead of coordinate descent for opt.knn (ablation) | opt.knn · iris convergence | MAE | No clear effect |
| increase K in opt.knn up to the sample size (ablation) | opt.knn · iris sensitivity | MAE | Mixed |
Closest papers by meaning
- Distribution Shift in Missing Data Imputation: A Risk-Based Perspective and Importance-Weighted Correction under MAR (NeurIPS 2026 · similarity 0.92)
- Learning What Not to Impute: An Uncertainty-Aware Diffusion Framework for Meaningful Missingness (NeurIPS 2026 · similarity 0.89)
- Random features models: a way to study the success of naive imputation (ICML 2024 · similarity 0.89)
- Unsupervised Anomaly Detection in The Presence of Missing Values (NeurIPS 2024 · similarity 0.89)
- DiffPuter: Empowering Diffusion Models for Missing Data Imputation (ICLR 2025 · similarity 0.88)
- Rethinking the Diffusion Models for Missing Data Imputation: A Gradient Flow Perspective (NeurIPS 2024 · similarity 0.88)
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.
