New SHAP-guided framework enhances LLM-based feature engineering

Researchers have developed SIGMA, a novel framework for automated feature engineering (AutoFE) that utilizes SHAP values to guide feature generation without requiring semantic metadata. This approach addresses limitations in existing LLM-based AutoFE methods, such as context window constraints and high duplicate feature rates. SIGMA's EXposed-feature Implicit Trajectory (EXIT) method maintains a constant prompt length and significantly reduces feature duplication from 37.2% to 6.8%, while achieving comparable performance to state-of-the-art methods with fewer features. AI

IMPACT This research offers a more efficient and scalable approach to feature engineering for LLMs, potentially improving performance in various machine learning applications.

RANK_REASON The cluster contains a research paper detailing a new methodology for LLM-based automated feature engineering. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• CatalyzeX
• DagsHub
• Hugging Face
• Large Language Models
• SHAP
• Shinichi Shirakawa
• SIGMA

• paper
• other

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