Model Validation & Refinement
Baseline vs candidate comparison with plain-language explanations. Improvements are simulated experiments, not proven results, and no button makes a model clinically approved.
Model comparison
HealthAI Solutions · Hospital readmission risk · candidate v2.3
AUROC
How well the model ranks higher-risk people above lower-risk people (0.5 = coin flip, 1 = perfect).
0.78
0.87
0.78 → 0.87 +9.0pt
Precision
Of the people the model flags, how many truly have the outcome.
0.68
0.76
0.68 → 0.76 +8.0pt
Recall / Sensitivity
Of the people who truly have the outcome, how many the model catches.
0.65
0.72
0.65 → 0.72 +7.0pt
Specificity
Of the people without the outcome, how many the model correctly leaves alone.
0.82
0.88
0.82 → 0.88 +6.0pt
Calibration error
Gap between predicted and observed risk — lower is better.
0.071
0.042
0.071 → 0.042 -2.9pt
False positive rate
Share of people flagged unnecessarily — lower is better.
0.18
0.12
0.18 → 0.12 -6.0pt
Summary radar
Baseline vs candidate
Executive read: the candidate ranks risk more accurately and flags fewer people unnecessarily — in this simulation.