# Purpose

When creating a baseline, you may need to exclude elements that correspond to abnormal situations. Indeed, taking into account abnormally high or low amounts would impact the baseline durably with irrelevant information.

Then, you would have to define a rule that will identify elements that are to be excluded of the computed baseline.

Example of a regular baseline vs filtered baseline

# Path to implement

To implement the filtered baseline it is necessary to identify what is an outlier. An outlier will impact durably the baseline by overestimating or underestimating the computed average.

## Filtered baseline with "simple" outlier definition

Example of definition:

When the measured count of item is greater than 10 times the computed average

The computation of the baseline would be as follows:

## Filtered baseline with baseline-based outlier definition

The outlier can be determined from a baseline-based evaluation. In this case, we will use 2 baselines:

• One baseline for outlier definition
• One filtered baseline for risk evaluation

The computation of the filtered baseline would be as follows:

# Steps (Filtered baseline with "simple" outlier definition)

#### 1 - Define a "baselinable" indicator rule

This step consists in defining a "smooth" value that removes abnormal situations.

In the example below, we assume that the count of transactions should not be greater than 10 * the average count over the last 40 days.

#### 2 - Create a "smooth" baseline

When creating the baseline, use the "baselinable count" indicator rather than the original count, so that the baseline will be filtered according to the rule defined in step 1.