LLM2Features: Large Language Models in Interpretable Feature Generation for AutoML with Tabular Data

Aleksandr Tsymbalov, Andrey Savchenko

Keywords: LLM, Auto feature generation, GPT-4o, GPT-o1, Tabular Data

TL;DR: our LLM2Features method generates features automatically from tabular data without any other additional actions from the user better than any other feature generation method. Only the table pd.DataFrame is enough

Abstract: Automatic Machine Learning (AutoML) is the popular supervised learning approach for tabular data. One of its key components is generating the most suitable features given the available training dataset. To overcome the disadvantages of existing automatic feature generation techniques, such as lack of generality and interpretability, we propose the novel approach, \textbf{LLM2Features}. It uses LLMs (Large Language Models) to generate meaningful features using automatically collected statistics about the dataset without explicitly describing the data, making it ideal for implementing in AutoML frameworks. In particular, we introduce the LLM-based critic that additionally verifies the presence of syntax or logical errors. The experimental study demonstrates the benefits of the proposed LLM2Features approach in accuracy and training time compared to the state-of-the-art feature generation tools.

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning

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.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

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: 8009