πŸ“Š Benchmark Dashboard

Zoning Opposition Prediction β€” Model Comparison

ROC Curve

Precision-Recall Curve

Calibration Diagram

Score Distribution

⚠️ Random 80/20 split β€” spatial autocorrelation inflates metrics

ROC-AUC by Eval Year

ECE by Eval Year (↓ better)

Lift@1% by Eval Year

Brier Score by Eval Year (↓ better)

Detailed Fold Results

βœ… Expanding-window temporal validation β€” no future data leakage. Scenario chaining active for hβ‰₯2.

Random Split vs Temporal β€” ROC-AUC by Model

Calibration Comparison (ECE ↓)

Brier Score Comparison (↓)

Key insight: Random-split AUC is inflated by 13-43pp due to spatial leakage. Temporal validation reveals true generalization.