Automated Feature Engineering by Prompting

Yufeng Zou, Jean Utke, Diego Klabjan, Han Liu

Keywords: Automated Feature Engineering, Large Language Models, Tabular Data Prediction

TL;DR: We propose a novel LLM-based AutoFE algorithm that leverages the semantic information of datasets.

Abstract:

Automated feature engineering (AutoFE) liberates data scientists from the burden
of manual feature construction, a critical step for tabular data prediction. While the
semantic information of datasets provides valuable context for feature engineering,
it has been underutilized in most existing works. In this paper, we introduce
AutoFE by Prompting (FEBP), a novel AutoFE algorithm that leverages large language
models (LLMs) to process dataset descriptions and automatically generate
features. Incorporating domain knowledge, the LLM iteratively refines feature
construction through in-context learning of top-performing example features and
provides semantic explanations. Our experiments on real-world datasets demonstrate
the superior performance of FEBP over state-of-the-art AuoFE methods. We
also conduct ablation study to verify the impact of dataset semantic information
and examine the behavior of our LLM-based feature search process.

Primary Area: applications to computer vision, audio, language, and other modalities

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 3279