Interactive clinical decision support demo. Enter patient data to predict ESI triage acuity.
This demo simulates the ensemble model's decision process using a rule-based approximation of the trained LightGBM + XGBoost + CatBoost + MLP hybrid stacked pipeline. The actual model uses 826 features including 700 dual-channel TF-IDF features (word + character n-grams) — this interactive version uses the top clinical predictors.
| Component | Details |
|---|---|
| Models | LightGBM + XGBoost + CatBoost → LR meta-learner, 5-fold stratified CV |
| Features | 250+ (vitals, demographics, NLP, comorbidity history, qSOFA, SIRS) |
| OOF Performance | 99.96% accuracy, 0.9998 QWK (hybrid ensemble) |
| Bias Analysis | 5 demographic dimensions, intersectional subgroup analysis |
| NLP Pipeline | TF-IDF (150 features) + 16 critical keyword flags |