Research map / Learning theory and algorithms / Kernels and regression theory
Data-Driven Subgroup Identification for Linear Regression
Zachary Izzo, Ruishan Liu, James Zou
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 |
|---|---|---|---|
| Use DDGroup region selection instead of whole-data linear regression | linear regression on real medical datasets · Brazil Health HF / stroke, China HIV, Dutch Drinking, Korea Grip (single group) | Test MSE: large drops on Brazil Health (0.80->0.04) and China HIV (0.84->0.69); small or none on China Glucose SUA-M and Korea Gri | Mixed |
| Restrict selected region to contain at least 5% of validation set (vs. unrestricted) | DDGroup · Real-world single group | Subgroup size and test MSE | No clear effect |
| Replace whole-data linear model with DDGroup subgroup (case study on blood transfusion route) | China HIV stigma · China HIV subgroup of 21% of participants | Coefficient and p-value for blood transfusion route: beta 0.12 (p=0.67) on whole data vs beta -1.71 (p=0.006) in subgroup | Improved |
| Replace DDGroup with k-means clustering for subgroup selection | single subgroup on real data · Real-world datasets | Test MSE: DDGroup generally lower MSE than clustering (e.g., Brazil HF 0.33 vs 0.04) | Hurt |
| Increase synthetic sample size for DDGroup (ablation) | DDGroup on synthetic data · Synthetic, n from 200 to 12800 | F1 score: 0.73 at n=200 to 1.00 at n=12800 | Improved |
| Increase synthetic sample size for k-means clustering | Clustering baseline · Synthetic, n from 200 to 12800 | F1 score: stays between 0.02 and 0.18 | No clear effect |
| Robustness of DDGroup to misspecified core group (ablation) | DDGroup core group initialization · Synthetic demonstration | Recall and F1: performance drops sharply once bad points enter the core group; precision stays above 0.22 baseline | Hurt |
| Select multiple subgroups iteratively (3 groups) with DDGroup | DDGroup · Real-world datasets, multiple groups | Test MSE: e.g., Brazil HF 0.80 baseline to 0.21 ± 0.03 | Improved |
Closest papers by meaning
- Data-IQ: Characterizing subgroups with heterogeneous outcomes in tabular data (NeurIPS 2022 · similarity 0.88)
- Dynamic Subgroup Identification in Covariate-adjusted Response-adaptive Randomization Experiments (NeurIPS 2024 · similarity 0.86)
- A Unified Framework for Variable Selection in Model-Based Clustering with Missing Not at Random (NeurIPS 2025 · similarity 0.84)
- Stochastic Grouping Conformal Prediction for Effective Subgroup Reliability (NeurIPS 2026 · similarity 0.84)
- Causal Representation Learning from Multimodal Biomedical Observations (ICLR 2025 · similarity 0.83)
- Dual Feature Reduction for the Sparse-group Lasso and its Adaptive Variant (ICML 2025 · similarity 0.83)
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.
