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?

More Anomaly Detection interview questions

View all →