Model Performance
Evaluate and monitor the real predictive performance of the demand-forecasting model across key metrics — every section below is either live data or clearly marked as not.
Accuracy by Horizon
Real weighted MAPE (%) across real horizon steps, NEM — GET /v1/forecast/recent-actual-vs-predicted
Loading real backtest…
Accuracy by Region
Real walk-forward MAPE (%), full horizon — this model vs. real seasonal-naive baseline
Loading…
Forecast Reliability (Calibration)
How well the real prediction intervals capture real actual demand
No coverage metric available.
Error Distribution
Real histogram of (actual − forecast) / actual, NEM, last 7 real days
Loading…
Model vs Benchmark
Real Production evaluation for each real architecture, plus the real seasonal-naive baseline
Loading real per-architecture evaluations…
Performance Over Time
Real avg walk-forward MAPE per evaluate run, lstm_demand
Loading…
Model Health Score
Real, computed from this page's own real inputs — formula disclosed below
Not enough real inputs yet.
35% accuracy (vs. naive baseline) + 15% calibration (coverage vs. target) + 15% drift (top-3 PSI) + 15% stability (MAPE variance across real evaluate runs) + 20% data quality (real 24h pass rate, GET /v1/data-quality/summary/public) — reweighted across whichever components have a real value.
Recent Alerts & Events
Built from this page's own real computed states — not sample text
Not enough real data yet to compute any event.