Newest papers

When Do LLMs Help with Feature Selection? A Systematic Empirical Study

Fateme Memar, Dongjie Wang, Arvin Agah · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.28556 · cited by 0 works (OpenAlex) · open-access PDF

Russian

Большие языковые модели (LLM) могут использовать семантические названия функций наряду с обобщающей статистикой для отбора табличных функций, но остается неясным, когда такая информация улучшает подбор подгрупп по отношению к установленным статистическим методам. Мы исследуем этот вопрос в рамках 18-ти наборов данных, состоящих из 10 наборов данных для групповой оптимизации относительной политики (GRPO) обучения, 2 для валидации контрольных пунктов, и 6 наборов данных для оценки по отдельности. Мы сравниваем пять классических базовых линий, замороженные селекторы Qwen2.5-7B и Llama-3.3-70B, и две модели Qwen2.5-7B, адаптированные к GRPO, сообщающие классификацию ROC-AUC и регрессию R^2 отдельно

Machine translation (Meta M2M100); the English original follows.

Abstract

Large language models (LLMs) can leverage semantic feature names alongside summary statistics for tabular feature selection, yet it remains unclear when such information improves subset selection relative to established statistical methods. We investigate this question within an 18-dataset framework comprising 10 datasets for Group Relative Policy Optimization (GRPO) training, 2 for checkpoint validation, and 6 held-out evaluation datasets. We compare five classical baselines, frozen Qwen2.5-7B and Llama-3.3-70B selectors, and two GRPO-adapted Qwen2.5-7B models, reporting classification ROC-AUC and regression R^2 separately and using dataset-wise ranks for cross-task comparison. Across five seeded evaluation partitions, classical selectors generally achieve higher predictive performance than the tested 7B LLM/RL selectors, including under a pre-specified ~33% feature budget with strict train-only classical fitting, while the 7B selectors retain compact subsets. A cardinality-matched semantic sensitivity experiment shows that, for Zero-shot Qwen-7B, informative feature names improve regression performance over anonymized and shuffled alternatives (+0.1964 and +0.1606 mean R^2), whereas aggregate classification differences are smaller (+0.0159 and +0.0084 ROC-AUC) and dataset-dependent. These results characterize when semantic information benefits LLM-based feature selection and show that classical selectors remain strong baselines for predictive tabular feature selection.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Feature selection, Leverage (statistics), Feature (linguistics), Computer science, Artificial intelligence, Regression

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange