[2602.11084] GRASP: group-Shapley feature selection for patients

Computer Science > Machine Learning arXiv:2602.11084 (cs) [Submitted on 11 Feb 2026 (v1), last revised 30 Apr 2026 (this version, v2)] Title:GRASP: group-Shapley feature selection for patients Authors:Yuheng Luo, Shuyan Li, Zhong Cao View a PDF of the paper titled GRASP: group-Shapley feature selection for patients, by Yuheng Luo and 2 other authors View PDF HTML (experimental) Abstract:Feature selection remains a major challenge in medical prediction, where existing approaches such as LASSO often lack robustness and interpretability. We introduce GRASP, a novel framework that couples Shapley value driven attribution with group $L_{21}$ regularization to extract compact and non-redundant feature sets. GRASP first distills group level importance scores from a pretrained tree model via SHAP, then enforces structured sparsity through group $L_{21}$ regularized logistic regression, yielding stable and interpretable selections. Extensive comparisons with LASSO, SHAP, and deep learning based methods show that GRASP consistently delivers comparable or superior predictive accuracy, while identifying fewer, less redundant, and more stable features. Comments: Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI) Cite as: arXiv:2602.11084 [cs.LG]   (or arXiv:2602.11084v2 [cs.LG] for this version)   https://doi.org/10.48550/arXiv.2602.11084 Focus to learn more arXiv-issued DOI via DataCite Submission history From: Zhong Cao [view email] [v1] Wed, 11 Feb 2026 17:50:57 UTC...