How to use the Outlier Finder.
Outliers depend on the chosen rule and context. A flagged value may be valid, so show the calculation and keep the original record.
Make the workflow fit your task.
Choose and state a detection rule before labeling observations. Show the threshold or calculation, retain row identifiers and review relevant groups separately. Treat flagged values as investigation candidates, checking units and source errors before excluding them.
- What you provide
- Numeric table and selected statistical rule.
- What you get
- Flagged observations with calculation and context cautions.
See the input and the result.
Illustrative input and output · a teaching example, not a live WebAct run
Example input
Values with row IDs: A=10, B=11, C=12, D=100. Use a supplied review rule: flag values above twice the median. Do not delete records.
Completed example
Sorted values: 10, 11, 12, 100. Median: (11 + 12) / 2 = 11.5. Threshold: 2 × 11.5 = 23. Flagged: row D, value 100, because 100 > 23. A, B and C do not exceed the threshold. This is the requested review heuristic, not evidence that D is an error.
Load this input into the prompt, then copy it to WebAct to try the task. Your result may differ from the illustration.
Decisions and troubleshooting.
Should every flagged outlier be removed before calculating an average?
No. An unusual value may be genuine and important. Document any exclusion rule and compare results with and without it.
Why does a valid high-value customer appear as an outlier?
The comparison population may combine different customer types. Examine segment context and the chosen detection threshold.
Try it with your own source.
Replace the example with your material in the task prompt. Keep the requirements you need, then copy the task into WebAct.
Customize and copy the task ↑