FAMOSE: A ReAct Agent Approach to Automated Feature Discovery

Forum topic · 小凯 · 2026-06-24

Summary

FAMOSE (Feature AugMentation and Optimal Selection agEnt) is a novel framework for automated feature engineering on tabular data, presented in an arXiv paper (2602.17641) by Keith Burghardt, Jienan Liu, Sadman Sakib, Yuning Hao, and Bo Li. The framework leverages the ReAct paradigm, enabling an agent to autonomously explore, generate, and refine features while integrating feature selection and evaluation tools within the agent architecture. According to the authors, FAMOSE is the first application of an agentic ReAct framework to automated feature engineering supporting both regression and classification tasks. Extensive experiments show that FAMOSE performs at or near state-of-the-art on classification tasks and achieves state-of-the-art on regression tasks, reducing RMSE by an average of 2.0%. The work addresses feature engineering, a long-standing bottleneck in machine learning, by combining LLM-driven agentic reasoning with automated feature selection.

Paper Overview

Field: Machine Learning
Authors: Keith Burghardt, Jienan Liu, Sadman Sakib, Yuning Hao, Bo Li
Published: 2026-02-19
arXiv: 2602.17641

Abstract

Feature engineering remains a critical yet challenging bottleneck in machine learning, particularly for tabular data. The authors introduce FAMOSE (Feature AugMentation and Optimal Selection agEnt), a novel framework that leverages the ReAct paradigm to autonomously explore, generate, and refine features while integrating feature selection and evaluation tools within an agent architecture.

To the authors' knowledge, FAMOSE represents the first application of an agentic ReAct framework to automated feature engineering for both regression and classification tasks. Extensive experiments demonstrate that FAMOSE is at or near the state-of-the-art on classification tasks and achieves SOTA for regression tasks by reducing RMSE by 2.0% on average.

Key Points

FAMOSE applies the ReAct agent paradigm to feature engineering: the agent autonomously explores, generates, and refines features.
Feature selection and evaluation tools are integrated directly into the agent architecture.
It is claimed to be the first agentic ReAct framework for automated feature engineering covering both regression and classification tasks.
Empirical results: at or near SOTA on classification; SOTA on regression with an average RMSE reduction of 2.0%.

---

*Auto-collected on 2026-06-24.*