KNN Model Comparison
Compare V20 (Direct SBP), V25 (MAP-First), V26 (30-Feature Expanded), and V27 (CT-on-V20-KNN) models
V20: Direct SBP KNN
15-feature KNN predicting SBP directly from upstream model estimates
Features
15 SBP
Architecture
Single-stage KNN
Status
Production
V25: MAP-First KNN
Two-stage KNN: predicts MAP first, then PP, derives SBP/DBP
Features
15 MAP + 15 PP
Architecture
Two-stage KNN
Status
Shadow
Backfill
Force
LOO
V26: Expanded 30-Feature KNN
30-feature KNN using 15 SBP + 15 MAP estimates for richer feature space
Features
30 (15 SBP + 15 MAP)
Architecture
Single-stage KNN
Status
Shadow
Backfill
Force
LOO
V27: CT-on-V20-KNN
KNN error correction applied to V20 KNN output for residual error reduction
Input
V20 KNN SBP/DBP
Architecture
KNN Error Corrector
Status
Shadow
Head-to-Head Comparison (LOO Evaluation)
Run Leave-One-Out evaluation on all three models simultaneously against the validated reference dataset
Run Full Comparison