The Outlier Detection feature helps you identify data points that are outliers in your charts. Amplitude calculates outliers by using an algorithm that first calculates the standard deviation of all data points on a chart. Then, the algorithm will analyze each data point and consider it an outlier if it is more than 1.5 standard deviations from any other data point. The outlier group will also be no greater than 0.1% of the overall population. For more information, you can read more about the DBSCAN data clustering algorithm here.
This feature allows you to create cohorts of users who are outliers in your product so that you can exclude them from your analyses or dig deeper into their behaviors.
Table of Contents
Note: Currently, outlier detection is only supported for the Event Segmentation chart type and is available to customers on our Enterprise Plan.
To calculate outliers in your chart, first set up your chart by picking the events and the metric you wish to calculate outliers for. If you select more than one event or have multiple segments, then this feature will detect outliers for each data set independently. After you have set up your chart, select '[Amplitude] Cohort' = '[Amplitude] Outliers' in the right module of the chart control panel. This will give you a segment that contains all users who are outliers. You can also add another segment of users who are not outliers. For example, let's say we wanted to find the group of users who are outliers when it comes to purchasing events in our product.
After you have enabled this feature, the chart will show you the data points that are outliers by grouping them into one segment. You will see a segment in the right module that represents your outliers.
Getting the Outliers
To get the users in the Outliers cohort, you can use the bar chart visualization to create a static Microscope cohort. You can then apply this cohort to other charts in Amplitude to further analyze these users to see why they are outliers.