A modular and interpretable framework for tabular data analysis using LLaMA 7B: Enhancing preprocessing, modeling, and explainability with local language models
Demonstrates a local LLM-assisted pipeline to enhance data preprocessing and explainability in tabular prediction tasks, indicating significant potential for healthcare efficiencies.
Key Points
The aim is to develop a framework using LLaMA 7B to improve data preprocessing and explainability in predicting medical appointment attendance.
Implemented a local LLM-assisted pipeline for semantic preprocessing.
Automated tasks included column renaming, datatype inference, and cleaning recommendations.
Applied the pipeline on the Medical Appointment No-Shows dataset.
Utilized XGBoost classifier for predictive modeling and SHAP for explainability.
Achieved an overall accuracy of 80% with the XGBoost classifier.
F1-score was 0.89 for the majority Show class and 0.03 for the minority No-show class.
AUC-ROC reached 0.65 and precision-recall AUC was 0.87, highlighting class imbalance effects.
Identified waiting days, age, and SMS notifications as key influential predictors through SHAP analysis.