A cycle is a repeating time-series pattern whose timing or length may vary. Unlike seasonality, which is more regular and calendar-linked, cycles can be less predictable, as with business cycles. section 17.2.3
Cycles change the context for outlier detection. A spike is unusual when it occurs rarely, but expected when it repeats regularly; missing an expected spike can itself be anomalous. section 17.2.3
Interpret a value against relevant historical cycles:
- Compare outdoor temperatures across daily and yearly patterns. section 17.2.3
- Compare network traffic across times of day and days of the week. section 17.2.3
- Compare sales with recent matching weekdays, month positions, and calendar periods from previous years. section 17.2.3
Removing a regular cycle
A time series with a spike every 10 seconds should normally contain a spike at 120 seconds. Subtracting the regular pattern exposes the residual values, where unusual behavior is easier to assess.
