Course contentsShow
← Data Science
Lesson 1410 of 2,14535. Statistical Anomaly DetectionFree lesson

Interpreting S-H-ESD Results

Reading anomaly scores, timestamps, and direction to understand which points deviate significantly from seasonal expectations.

Interpreting S-H-ESD Results

What you'll learn: How to read and make sense of Seasonal Hybrid ESD outputs to identify which data points are genuine anomalies.

Understanding the Output

When you run S-H-ESD, you don't just get a simple "yes/no" answer. The algorithm returns several pieces of information for each detected anomaly:

Anomaly Scores

These numerical values indicate how unusual each flagged point is relative to the expected seasonal pattern. Higher scores mean greater deviation. Think of it like a "weirdness meter"—a score of 3 is mildly odd, but a score of 10 is genuinely shocking.

Timestamps

Each anomaly is tagged with its exact position in your time series. This lets you pinpoint when the unusual behavior occurred, which is critical for root-cause analysis. Was it during a holiday? A system outage? A marketing campaign?

Direction (Positive or Negative)

S-H-ESD tells you whether the anomaly is a spike (above expected values) or a dip (below expected values). This matters tremendously: a sudden traffic spike might indicate viral growth or a DDoS attack, while a dip could signal system failure or lost data.

Practical Interpretation

When reviewing results, prioritize anomalies with the highest scores first. Cross-reference timestamps with business events or system logs. The direction helps you understand the nature of the problem—spikes aren't always bad (sales surge!), and dips aren't always problems (planned maintenance).

Remember: S-H-ESD accounts for seasonal expectations, so a "normal" peak during high season won't trigger false alarms, but an unusually high peak will.

Key Takeaway: S-H-ESD results give you anomaly scores (magnitude of deviation), timestamps (when it happened), and direction (spike or dip)—together, these three elements help you prioritize investigation and understand what actually went wrong or right in your data.