Real-Time Anomaly Detection

Live sensor stream (temperature, pressure, vibration) with online anomaly detection. The model updates with every observation and resets when the distribution shifts (concept drift).

Disconnected
0
0
0.00
0.00

Last 200 observations. Red dots = anomaly alerts; yellow vertical lines = drift (model reset).

What you're seeing
  • Synthetic stream: Three phases — normal baseline, gradual concept drift, then injected anomalies (spikes and contextual outliers). The stream loops; each cycle takes ~40s.
  • Online learning: The detector updates with every observation; no batch retraining. When drift is detected on the anomaly score stream, the model resets to adapt.
  • Metrics: Precision and recall are computed in real time against ground truth labels. All counters reset at the start of each stream cycle so the dashboard reflects the current run.