Free browser-based ML tool

Outlier Calculator

Find outliers in CSV data with IQR, Z-score or modified Z-score. Review original rows, export flags and charts, and reproduce the calculation locally.

Free to useNo upload or sign-inRuns entirely in your browser
Live calculationPowered by @kanaries/ml

Data stays in this tab. Limits: 2,000 rows, 30 columns; customer clustering: 500 customers and 12 features.

Choose columns, then calculate. Changing an input hides the previous result until you run again.

Preview input (10 rows)
record_idvalue
00110
00211
00312
00412
00513
00614
00715
00816
00917
01080

Review unusual observations without losing context

Paste a CSV with a header row, select a numeric column and calculate. Text IDs and other fields stay attached to each observation. Switch the detection rule to compare how assumptions change the flagged rows. The chart highlights flagged points in red; the export keeps both flagged and unflagged records.

Understand the default example

The sample contains nine values between 10 and 17 and one value of 80. Linear-interpolated quartiles are 12 and 15.75, producing IQR fences of 6.375 and 21.375. IQR flags 80. The strict population Z-score rule does not flag it because its score remains below 3. This is a useful demonstration of how a large value changes its own reference mean and spread.

Choose the unit of analysis

A single-column rule cannot identify every unusual combination of features. Review comparable populations together and keep units consistent. The JavaScript Isolation Forest guide covers a multivariate alternative. A statistical flag is not a diagnosis of fraud or proof of a data error.

Work through a practical example

How to Find Outliers in a CSV File Without Python explains the input, assumptions and expected output. Continue with IQR vs Z-Score vs Isolation Forest: Compare the Same Dataset to compare methods and interpret results.

Frequently asked questions

Questions about outlier calculator

Which rules are available?

IQR uses 1.5 times the interquartile range with linear-interpolated quartiles. Z-score uses population standard deviation and an absolute threshold above 3. Modified Z-score uses median absolute deviation and a threshold above 3.5.

Does a flagged row need to be removed?

No. Original records remain in the results. Investigate units, data quality and the population before deciding whether to correct, retain or exclude a value.

What if all values are identical?

IQR flags none. Z-scores are undefined with zero standard deviation, and modified Z-scores are undefined with zero MAD. The tool explains these cases.

Is my CSV uploaded?

No. Parsing, fitting and chart rendering run in this browser tab. Files are limited to 1 MB, 2,000 rows and 30 columns; clustering also limits the input to 500 customers and 12 features.

Powered by @kanaries/ml — the scikit-learn-style ML library for JavaScriptnpmGitHub