auphong2707/hospital-readmission-phase5-lgbm-fairness

Hospital Readmission Risk - Phase 5: Fairness Evaluation Results

This repository contains the results from Phase 5: Fairness Evaluation & Deployment Readiness.

Contents

Outputs

• outputs/fairness_report.json: Comprehensive fairness evaluation report
• outputs/group_metrics_*.csv: Performance metrics by demographic group (race, gender, age)
• outputs/statistical_tests.json: Statistical significance tests for bias detection
• outputs/risk_categories_*.csv: Risk category distribution by demographic group

Fairness Metrics Evaluated

Demographic Parity

Measures if intervention rate is similar across demographic groups (±5% tolerance).

Equalized Odds

Measures if True Positive Rate (TPR) and False Positive Rate (FPR) are similar across groups (±5% tolerance).

Equal Opportunity

Measures if True Positive Rate (TPR) is similar across groups (±5% tolerance).

Statistical Tests

• Chi-square test: Tests independence of intervention rate and demographic group
• Two-proportion z-test: Tests TPR/FPR differences between groups

Model Information

• Model: Gradient Boosting (LightGBM) with Platt Calibration
• Optimal Threshold: From Phase 4 ROI analysis
• Test Set: 15,265 patients
• Demographics: Race (6 categories), Gender (3 categories), Age (10 ranges)

Usage

These results can be used for:

• Assessing model fairness before deployment
• Identifying potential bias in predictions
• Determining if bias mitigation is needed
• Creating model cards with fairness documentation
• Meeting regulatory requirements for AI fairness

Deployment Readiness

Review the fairness report to determine if the model is ready for deployment or if bias mitigation strategies are needed.

Citation

If you use these results, please cite the hospital readmission risk prediction project.