Create intermediate Works with: Any AI

Outlier Handling Decision Guide

Decision tree Has example

Decide to keep, fix, or remove extreme values.

The Open buttons send the prompt text as currently shown below to that site only when you click. Fill in the placeholders first.

Help me decide what to do with outliers. Variable and context: [VARIABLE_AND_CONTEXT]. Summary and the extreme values: [SUMMARY_AND_EXTREMES]. What the analysis will be used for: [USE_CASE].

Use this decision tree and show the path:
1. Is the value impossible (negative age, a date in the future)? If yes, treat it as an error: correct from the source, set to missing, or exclude, and record it.
2. Is it possible but suspicious (a unit error such as grams versus kilograms, a typo with an extra zero)? Look for a pattern and test the hypothesis.
3. Is it real and rare (a large customer, a record sale)? Keep it, and consider robust methods (median, trimmed mean, winsorizing, log transform) or reporting results with and without it.
4. Does it drive the conclusion? Run a sensitivity analysis.

Provide the detection code (IQR rule: below Q1 - 1.5 x IQR or above Q3 + 1.5 x IQR; z-scores; domain limits) with a warning that these rules flag values, not errors. Output a table with Value | Flag reason | Classification | Decision | Justification, and a sentence for the methods section describing what was done. Never remove a point only because it spoils the result.

Fill in before use

Optional: type your details and the prompt above updates. Nothing is saved or sent.

Replace each bracketed placeholder with your own details, then paste into Any AI or another assistant.

Data: delivery times in minutes; most 20-60, a handful over 600.
Questions: Are they errors (typed 6000 instead of 60)? Rare but real events (storms)? Different process (custom orders)?
Decision table
Cause | Action | Document
Data entry error | Correct or set to missing | Rule and count
Real extreme | Keep; use robust summaries (median, trimmed mean) | Note
Different population | Analyse separately | Reason
Never delete values just because they hurt your result.
Sensitivity: report results with and without the 7 extreme rows.
Plot: boxplot plus a histogram on log scale.

• outliers