Demand Intelligence / M1

Forecast Performance & Model Monitoring

Backtesting, validation, monitoring and rule-based alerts for the demand forecast.

DEMO performance metrics — not a validated AI model. No model has been trained or connected. Accuracy, errors, coverage and alerts are computed from synthetic actuals and synthetic predictions to demonstrate the monitoring workflow, and must not be read as evidence of real forecast accuracy.

Forecast Accuracy

95.8%

100 − MAPEDEMO

Mean Absolute Error (MAE)

410

vehicle entries / dayDEMO

MAPE

4.2%

84 forecast–actual pairsDEMO

Forecast Bias

-1.3%

under-forecastingDEMO

Model Status

Needs attention

8 rule-based alertsDEMO

Actual vs. Predicted DemandDEMO

Daily entries · 1 day-ahead predictions · last 28 days

Actual (synthetic)Predicted (synthetic)90% prediction interval

Forecast Error TrendDEMO

Daily % error (bars) and 7-day rolling MAPE (line)

Accuracy by FacilityDEMO

Accuracy threshold 90% (DEMO rule)

Facility	Accuracy	MAE	MAPE	Bias	Interval coverage	Recent 7-day MAPE
Terminal Parking A Short stay · Drive-up	94.7%	218	5.3%	-2.0%	86.9%	5.7%
Terminal Parking B Short stay · Mixed	95.4%	171	4.6%	-0.5%	94.0%	5.1%
Long Stay Parking Long stay · Pre-book	95.0%	59	5.0%	-1.2%	84.5%	5.6%
Premium Parking Premium · Mixed	95.4%	36	4.6%	-1.4%	88.1%	4.4%

Accuracy by Forecast HorizonDEMO

Error grows with lead time

Confidence Interval CoverageDEMO

Measured share of actuals inside the nominal 90% interval — different from the stated 'Forecast Confidence' width on Demand Forecasting

Model MonitoringDEMO

Model version

DEMO-M1-DF v0.3.2 (synthetic)

Last evaluation

13 Oct 2026, 23:00 AST

Last training date

29 Sep 2026, 02:00 AST

Data freshness

106 min (oldest feed)

Model drift status

Early signs · 1.22×

Retraining status

Recommended — requires RAC approval

Synthetic monitoring records only · no retraining is triggered automatically

Model AlertsDEMO

Data qualityinfo

Terminal Parking A: last synthetic record 106 min old (>90 min)

Data qualitywarning

Long Stay Parking: input completeness 94.9% below 97%

Data qualityinfo

Long Stay Parking: last synthetic record 103 min old (>90 min)

Missing input datacritical

Long Stay Parking: no input records for 11 Oct

Data qualitywarning

Premium Parking: input completeness 91.3% below 97%

Missing input datacritical

Premium Parking: no input records for 02 Oct

Missing input datacritical

Premium Parking: no input records for 11 Oct

Forecast anomalyinfo

Forecast anomaly 12 Oct: error -10.5% (z = -2.4)

Rules: accuracy <90% · drift >1.3× · completeness <97% · freshness >90 min · missing day · coverage <85% · |z| residual outliers

Forecast ValidationDEMO

Synthetic validation — not evidence of actual model accuracy

Historical validation periods (backtest folds)

Period	MAPE	Bias	Coverage
16 Sept – 22 Sept	4.1%	+0.8%	100.0%
23 Sept – 29 Sept	3.0%	+1.1%	100.0%
30 Sept – 06 Oct	4.8%	-4.4%	95.2%
07 Oct – 13 Oct	4.8%	-2.4%	85.7%

Baseline model comparison (MAPE)

DEMO forecast (selected horizon)	4.2%
Seasonal naive (same weekday, prior week)	4.9%
7-day moving average	8.7%

Performance by parking product

Product	Accuracy	MAPE	Coverage
Short stay	95.7%	4.3%	95.2%
Long stay	95.0%	5.0%	84.5%
Premium	95.4%	4.6%	88.1%

Error distribution (% error, all pairs)

Validation results (DEMO acceptance checks)

Accuracy ≥ 90%	95.8%	Pass (demo)
Beats seasonal-naive baseline (MAPE)	4.2% vs 4.9%	Pass (demo)
Beats 7-day moving-average baseline	4.2% vs 8.7%	Pass (demo)
90% interval coverage ≥ 85%	95.2%	Pass (demo)
\|Bias\| ≤ 3%	-1.3%	Pass (demo)

Acceptance thresholds are placeholders pending RAC-defined model governance criteria.