MedFeat uses LLMs for targeted clinical data feature engineering

Researchers have developed MedFeat, a novel framework for feature engineering in clinical tabular prediction tasks. This system uses Large Language Models (LLMs) to generate feature transformations, but unlike previous methods, it incorporates model-awareness and feature importance signals. This allows MedFeat to iteratively guide feature discovery, specifically tailoring proposals to the downstream model's needs and the characteristics of healthcare data, such as class imbalance and interpretability requirements. Evaluations show MedFeat significantly outperforms existing baselines, achieving over a 10% average improvement across various clinical datasets and models. AI

IMPACT Enhances clinical prediction accuracy by enabling LLMs to generate more effective features tailored to specific models and data challenges.

RANK_REASON The cluster contains a research paper detailing a new method for feature engineering using LLMs in a clinical context. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• clinical tabular prediction
• LLMs
• MedFeat
• Zizheng Zhang

• paper
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →