LLMs for Semi-Automated Data Science: Introducing CAAFE for Context-Aware Automated Feature Engineering Anonymous Author(s) Affiliation Address email

Keep change if performance is improved. Tabular Prediction User: Specifies LLM: Generates Interpreter: Model: problem context Code for feature Executes Performs and dataset engineering generated code cross-validation. Figure 1: CAAFE accepts a dataset as well as user-specified context information and operates by iteratively proposing and evaluating feature engineering operations.