Evolving LLM-Generated Features for Interpretable Classification

Abstract

Large language models (LLMs) are increasingly used as classifiers, yet they operate as opaque systems whose decisions are difficult to interpret, which complicates their use in regulated domains such as credit scoring or medical diagnosis. We propose an evolutionary framework that iteratively discovers natural language feature definitions (rubrics) for interpretable classification. An LLM generates candidate binary features, evaluates each sample against them, and the resulting vectors can be used to train a transparent classifier such as logistic regression. The feature set evolves over multiple iterations guided by classification errors, per-class activation rates, and feature ablation scores. We evaluate across three benchmarks, including a credit risk dataset representative of regulated domains, comparing single-shot LLM rubrics, evolved rubrics, and direct zero-shot LLM classification. Evolved features improve over single-shot rubrics by +2.9 pp on average and outperform zero-shot LLM classification on two of three tasks, while providing fully auditable decision logic. On the credit risk task, the zero-shot LLM performs at chance (50.7%) with a strong bias toward a single class, whereas evolved features achieve balanced, interpretable predictions. Crucially, this failure is invisible in aggregate accuracy and surfaces only under per-class auditing. Analysis reveals that evolution is most effective when label boundaries cannot be inferred from category names alone or when the LLM lacks reliable domain-specific reasoning.

Flat Loss, Evolving Features: Grokking in Modular Addition
Akash Kumar, Karthik Sankaranarayanan
arXiv15 Oct 2026PreprintCC BY 4.0Full textStatisticsarXiv
Classification and C₂-cofiniteness of rational vertex operator algebras generated in low conformal weights
Jiancai Sun, Ruochen Yang
arXiv4 Nov 2026PreprintarXiv licenceFull textAlgebra & number theoryarXiv
Coxeter Rigidity and ADE Classification of Fibonacci-Generated Tensor Categories
Huixuan He, Zhengwei Liu
arXiv30 Oct 2026PreprintarXiv licenceFull textAlgebra & number theoryarXiv
Clinically Interpretable Biological Age Estimation Using CatBoost and SHAP with Continuous and Categorical Features from NHANES
Yanni Sun, Zheng Qu, Kevin Ren +1
arXiv14 Oct 2026PreprintarXiv licenceFull textStatisticsarXiv
Interpretable EOG-based gaze classification using tree-based models: A SHAP and LIME analysis
Ayse Nur Ay Gul
PLOS One29 Oct 2026Peer-reviewedCC BYFull textGeneral biology Hydrology & soilsPLOSEurope PMCPMC

Cite

BibTeX RIS

BibTeX

@misc{butler2026evolving,
  title = {{Evolving LLM-Generated Features for Interpretable Classification}},
  author = {Jack Butler and Zainab Afolabi and Nikita Kozodoi},
  howpublished = {arXiv},
  year = {2026},
  month = oct,
  doi = {10.48550/arxiv.2610.03951},
  eprint = {2610.03951},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2610.03951},
  note = {CC BY 4.0},
}

RIS

TY  - UNPB
TI  - Evolving LLM-Generated Features for Interpretable Classification
AU  - Jack Butler
AU  - Zainab Afolabi
AU  - Nikita Kozodoi
PY  - 2026
DA  - 2026/10/02
PB  - arXiv
DO  - 10.48550/arxiv.2610.03951
UR  - https://arxiv.org/abs/2610.03951
AB  - Large language models (LLMs) are increasingly used as classifiers, yet they operate as opaque systems whose decisions are difficult to interpret, which complicates their use in regulated domains such as credit scoring or medical diagnosis. We propose an evolutionary framework that iteratively discovers natural language feature definitions (rubrics) for interpretable classification. An LLM generates candidate binary features, evaluates each sample against them, and the resulting vectors can be used to train a transparent classifier such as logistic regression. The feature set evolves over multiple iterations guided by classification errors, per-class activation rates, and feature ablation scores. We evaluate across three benchmarks, including a credit risk dataset representative of regulated domains, comparing single-shot LLM rubrics, evolved rubrics, and direct zero-shot LLM classification. Evolved features improve over single-shot rubrics by +2.9 pp on average and outperform zero-shot LLM classification on two of three tasks, while providing fully auditable decision logic. On the credit risk task, the zero-shot LLM performs at chance (50.7%) with a strong bias toward a single class, whereas evolved features achieve balanced, interpretable predictions. Crucially, this failure is invisible in aggregate accuracy and surfaces only under per-class auditing. Analysis reveals that evolution is most effective when label boundaries cannot be inferred from category names alone or when the LLM lacks reliable domain-specific reasoning.
ER  -

Abstract

Related

Cite