Anomaly Detection
Validation: February 9 Event
Detection results per method:
Matrix ProfileIsolation ForestLstm AutoencoderSpcTsfresh
The Feb 8-9, 2026 event is a clinically confirmed acute episode (HR spike to 113 bpm mean, RMSSD drop to 5.0 ms, readiness score 29). All anomaly detectors are validated against this event.
Anomaly Timeline
Statistical Process Control (SPC)
LSTM Autoencoder - Training Curve
Top Anomalies (Ensemble Scoring)
| # | Date | Ensemble Score | RMSSD (ms) | HR (bpm) | Efficiency | Readiness | Methods (score > 0.5) |
|---|---|---|---|---|---|---|---|
| 1 | 2026-07-11 | 0.934 | 8.0 | 106 | 58% | 39 | 4/5 |
| 2 | 2026-07-10 | 0.902 | 19.8 | 130 | 78% | 75 | 4/5 |
| 3 | 2026-02-09 ** | 0.873 | 5.0 | 110 | 76% | 29 | 4/5 |
| 4 | 2026-05-31 | 0.850 | 35.9 | 83 | -% | 63 | 4/5 |
| 5 | 2026-04-08 | 0.791 | 9.1 | 112 | 67% | 50 | 5/5 |
| 6 | 2026-02-10 | 0.783 | 13.0 | 94 | -% | - | 4/5 |
| 7 | 2026-06-04 | 0.774 | 26.0 | 84 | 58% | 65 | 4/5 |
| 8 | 2026-05-30 | 0.766 | 23.5 | 91 | 73% | 64 | 4/5 |
| 9 | 2026-06-02 | 0.752 | 30.2 | 81 | 75% | 66 | 3/5 |
| 10 | 2026-08-23 | 0.749 | 22.8 | 94 | 68% | 46 | 2/5 |
| 11 | 2026-07-17 | 0.747 | 30.7 | 100 | 84% | 70 | 3/5 |
| 12 | 2026-09-13 | 0.739 | 17.7 | 76 | 68% | 47 | 3/5 |
| 13 | 2026-06-01 | 0.738 | 39.2 | 87 | 84% | 80 | 3/5 |
| 13 | 2026-06-01 | 0.738 | 39.2 | 87 | 82% | 80 | 3/5 |
| 14 | 2026-08-25 | 0.736 | 35.8 | 83 | 88% | 68 | 4/5 |
| 15 | 2026-05-21 | 0.733 | 25.4 | 98 | 88% | 60 | 3/5 |
** = known acute event. Methods (score > 0.5) is a soft support count and is not the same as strict Feb 9 detection validation above.
Clinical Interpretation
Summary
Five independent anomaly detection algorithms were run retrospectively on 292 days of biometric data from the Oura Ring (Jan-Mar 2026).
Validation Against Known Event
3 of 5 methods detected the acute event on Feb 8-9, 2026. The event ranks as #3 in ensemble anomaly scoring.
This event showed dramatic HRV collapse (RMSSD 5.0 ms), mean heart rate 113.9 bpm during sleep, and readiness score 29/100 with recovery index 4/100 - consistent with acute autonomic destabilization.
Top 5 Anomalies
- 2026-07-11
- 2026-07-10
- 2026-02-09
- 2026-05-31
- 2026-04-08
Clinical Relevance
Baseline RMSSD ~10 ms is critically low (parasympathetic deficiency threshold (Kleiger 1987 / Bigger 1992): 15 ms, population median: 49 ms, typical post-HSCT: 25-40 ms). Any further decline represents severe autonomic stress. Mean nightly heart rate >89 bpm meets IST criteria (>90 bpm at rest).
Limitations
- Only 1 known positive event for validation (no specificity/PPV calculation possible)
- LSTM autoencoder has limited training data (292 days) - results should be interpreted cautiously
- Oura Ring measurements are not medical devices and have inherent noise
- tsfresh analysis depends on the number of HR data points per night (varies significantly)
Method Summary
| Method | Runtime | Feb 9 Detected | Details |
|---|---|---|---|
| Matrix Profile | 15.24s | No | 6 signals x windows |
| Isolation Forest | 0.33s | Yes | 30 anomalies / 292 days |
| Lstm Autoencoder | 6.3s | Yes | 22 anomalies, threshold=0.5586 |
| Spc | 0.01s | Yes | Shewhart + CUSUM + EWMA |
| Tsfresh | 0.29s | No | 10 features, 1 outliers |