How can you effectively monitor metrics with cyclic patterns using detectors?

Question

Options

• AApply a moving average to smooth out cyclic fluctuations
• BConfigure multiple detectors with different extrapolation policies
• CUse historical data to predict cyclic patterns
• DTrigger alerts based on the highest and lowest points of the cycles

• A
  76% (29)
• B
  8% (3)
• C
  3% (1)
• D
  13% (5)

Explanation

Applying a moving average (option A) is the right approach because it smooths out the regular up-and-down swings of cyclic data, exposing the underlying trend and making anomalies stand out against the expected pattern - without the noise of normal cycles triggering false alerts.

Why the distractors are wrong:

• B - Multiple detectors with different extrapolation policies addresses forecasting strategy, not the cyclic smoothing problem itself.
• C - Using historical data to predict cycles describes forecasting/seasonality modeling, which is a separate concern from real-time detection; prediction alone doesn't help a detector distinguish a genuine anomaly from a normal peak.
• D - Triggering on highest/lowest cycle points would fire alerts on every normal cycle, creating constant noise rather than meaningful signals.

Memory tip: Think of a moving average as a "trend filter" - it lets the cyclic waves pass through while the detector watches the smoothed line. The phrase "smooth to detect" can help you recall that cyclic patterns need smoothing before alerting.

No community discussion yet for this question.