CoFEH: LLM-driven Feature Engineering Empowered by Collaborative Bayesian Hyperparameter Optimization
Researchers have developed CoFEH, a novel framework that integrates Large Language Models (LLMs) with Bayesian Hyperparameter Optimization (HPO) for end-to-end automated machine learning. This system uses an LLM with a Tree of Thought approach to generate flexible feature engineering pipelines and a Bayesian optimization module for HPO. CoFEH uniquely interleaves these processes, allowing for informed decision-making between feature engineering and hyperparameter tuning, which has shown superior performance compared to existing methods. AI
IMPACT This framework could streamline the development of machine learning models by automating complex feature engineering and hyperparameter tuning processes.