UC-002_predictive_maint_model_test_results_v1.xlsx
EV-2026-0146 · Test Results · linked to Predictive maintenance — water mains
Artefact ID
EV-2026-0146
Type
Test Results
Linked use case
Uploaded by
Linda Park
Uploaded at
2026-05-25 11:08
Size
1248 KB
SHA-256
5e2c…a87b (truncated for display)
Hash chain
Verified · chained to previous entry
Summary
Holdout test set results for the predictive maintenance XGBoost model. Tested on final 12 months of historical data not seen during training. Meets AIRC condition (FP < 15%).
Chain of custody
Generated by Linda Park via Azure ML notebook 'pred-maint-eval-v1'. Hash chained on upload.
Key facts
- Test window
- 1 May 2025 – 30 April 2026
- Events tested
- 2,418 (including 187 confirmed failures)
- Suburbs covered
- All 9 pilot zone suburbs (Westbrook CBD)
- Model version
- xgb-v1.0.4 (training run 2026-05-18 17:42 AEST)
Headline metrics — test set (12 months holdout, 2,418 events)
| Metric | Value | AIRC threshold | Pass / fail |
|---|---|---|---|
| F1 score | 0.71 | ≥ 0.65 | ✓ Pass |
| Precision | 0.74 | ≥ 0.70 | ✓ Pass |
| Recall | 0.68 | ≥ 0.60 | ✓ Pass |
| False positive rate | 11.2% | < 15% | ✓ Pass |
| False negative rate | 8.4% | < 20% | ✓ Pass |
| Lead time (median, days) | 27 | ≥ 21 | ✓ Pass |