How do windowing techniques work in Time-Series Anomaly Detection?
Updated May 5, 2026
Short answer
Analyzing data in small overlapping chunks[cite: 1].
Deep explanation
Instead of global stats, we look at the last N points. This captures local trends and seasonality[cite: 1].
Real-world example
Flagging a sudden spike in CPU usage compared to the last 5 minutes[cite: 1].
Common mistakes
- Choosing a window size that is too small (noise) or too large (missing quick spikes)[cite: 1].
Follow-up questions
- Overlap vs. Tumble?