[Python] Anomaly Detection: Identifying Irregular Signal Patterns in Time Series

I have a time series dataset and need to detect anomalous patterns highlighted in red boxes in the attached plots.

Normal vs Anomalous Signals


Normal patterns:

• Baseline: Stable oscillations around zero
• Valid spikes: Clean amplitude transitions (>1000)

Anomalous patterns (target for detection):

• Irregular, chaotic fluctuations with high-frequency noise
• Can occur in baseline regions OR cross through valid signal areas
• Distinguished by noisy, non-smooth morphology compared to clean transitions

Challenge


The anomalies can span across multiple signal regions, making detection more complex. Like attach 2.

Questions

1. What algorithms work best for morphological anomaly detection?
2. How to distinguish valid spikes from irregular patterns that may cross signal boundaries?
3. Recommended Python libraries?

Data: Sample datasets: Dataset 1 Dataset 2



Continue reading...