Anomaly Detection in Usage Data

CELUS uses the Interquantile Range method to identify anomalies in usage data. 

To determine whether the value for a given month is unusual, CELUS looks back at the 12 preceding months of data for the same metric, platform, organization, and report type.

Let’s illustrate the process with an example.

Suppose we want to check whether Denials metric was unusual in January 2025 for a given platform and report type. To do this, CELUS examines the Denials values from the preceding 12 months – that is, from January 2024 to December 2024. You can always view these historical values in the chart displayed when you expand a row in the anomalies table.

Let’s say these are the Denials values from January 2024 to December 2024:

The value for January 2025 is 2367.

Defining Anomalies

We calculate quartiles (see Wikipedia) from the historical data.

• The median (Q2) is the middle value of the sorted data.
• The first quartile (Q1) is the point below which 25% of the data lies.
• The third quartile (Q3) marks the point below which 75% of the data lies.

For our 12 data points, the median is the value between the 6^th and 7^th entries, which is equal to 761.5. The first quartile lies between the 3^rd and 4^th value (731.5) and the third quartile between the 9^th and 10^th (766.5).  This calculation is illustrated above.

From Q1 and Q3 we compute the Interquantile Range as IQR = Q3 – Q1 = 766.5 – 731.5 = 35. This IQR is then used to determine the upper and lower boundaries of the range in which values are considered typical:

Lower boundary = Q1 – 5 × IQR = 555.5

Upper boundary = Q3 + 5 × IQR = 941

This band — the interval between the lower and upper boundaries — represents the range where values are expected to fall. It is also visualized in the chart, together with the median and both boundaries, to help you understand why a value is considered an anomaly.

Since our January value (2367) lies outside this band, CELUS flags it as an anomaly.

Traditional statistical methods typically use 1.5 × IQR to identify outliers. CELUS uses a more conservative multiplier of 5 × IQR to focus only on truly significant deviations.

Understanding Significance

Not all anomalies are equally important — some represent mild deviations, while others are extreme. To help you distinguish between them, CELUS calculates a metric called Significance for each anomaly.

Significance expresses how many IQRs the current value lies beyond the normal boundaries. In other words, it shows how far the anomaly is from what’s typical, relative to the usual variability in the data.

CELUS classifies anomalies based on their significance value:

Low significance (<= 10),

Medium significance (<= 25)

High significance (>25).

In our example, the distance of the anomaly from the upper boundary is equal to 1426. This distance can be expressed in terms of IQRs as approximately 41 × IQR, giving this anomaly its significance of 41 – a highly significant deviation.