Code and Processed Data for A Fairness-Aware and Explainable Machine Learning Framework for Predicting 30-Day Hospital Readmission in Patients with Diabetes

Abstract

This Jupyter Notebook presents a comprehensive pipeline for predicting hospital readmissions in diabetic patients using the "Diabetes 130-US hospitals" dataset. The study focuses on the dual challenge of model performance and ethical AI. Key features of this notebook: Preprocessing: Handling of missing data and categorical encoding for clinical variables. Modeling: Implementation of XGBoost as the primary predictive model. Fairness Analysis: Utilization of the Fairlearn library to evaluate and mitigate demographic parity and equalized odds issues across sensitive attributes like race and gender. Explainability (XAI): Application of SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations) to provide global and local interpretations of the model's decisions. This resource is intended for researchers and practitioners interested in Algorithmic Fairness and Interpretable Machine Learning in healthcare.