Outlier Detector

Set up outlier detection

1. Drop a CSV, TSV, JSON, JSONL, or spreadsheet file into the upload area, or click browse files.
2. Under Columns to analyze, keep at least one detected numeric column selected. A column appears when at least 60% of its present, non-missing values are numeric.
3. Choose a Method and enter a Threshold. The defaults are 1.5 for IQR, 3 for Z-score, and 3.5 for Modified Z-score. The threshold must be positive.
4. Choose Group by (optional) to calculate statistics separately for each group value. Choose Row identifier when a source column should label each result.
5. Review the table, then select Export CSV or Export JSON to download the flagged results.

How IQR, Z-score, and Modified Z-score find outliers

IQR calculates quartiles and flags values beyond the lower or upper bound. Z-score uses the mean and sample standard deviation. Modified Z-score uses the median, median absolute deviation, and the constant 0.6745. A value is flagged when its score reaches the selected threshold.

With IQR and the default threshold, the column below produces one flagged result:

amount
12
15
18
21
999

Row ID  Column  Value  Method  Threshold  Score  Severity
5       amount  999    IQR     1.5        161.5  extreme

Flagged results are Mild below a score-to-threshold ratio of 1.5, Moderate from 1.5 up to 2.5, and Extreme at 2.5 or higher.

Find outliers in numeric columns with groups

Use Group by (optional) when each group should have its own statistics. A missing group value appears as (missing). The selected row label comes from Row identifier. If that value is missing, the result uses row N; Row number uses the row number.

Blank, missing, and non-numeric values in selected columns are excluded and counted separately. The tool does not convert them to zero. A column or group is skipped when it has fewer than three valid values, or when the selected method has zero IQR, standard deviation, or median absolute deviation.

What the outlier report contains

Each flagged row shows Row ID, Column, Group, Value, Method, Threshold, Score, and Severity. Rows are sorted by descending score. The summary counts analyzed values, excluded values, flagged records, severity levels, and results by column.

The table displays the first 500 flagged records. Export CSV and Export JSON include every flagged record, with result fields rather than full source rows.

File formats and parsing rules

• CSV and .txt files are comma-delimited; TSV files use tabs. JSON must contain an array at the root or inside an object property.
• JSONL and .ndjson use one JSON object per non-empty line. XLSX and XLS files use the first worksheet. Empty or unreadable files produce an error.

The file is read and scored in the browser. This page's code does not send the selected file contents in a request. The tool never modifies the source data.

Keep groups and row labels in the report

Choose a column under Group by (optional) when its values should define separate calculations. Choose a source column under Row identifier when the report should keep that column's labels instead of row numbers.

Choose a dataset analysis tool

Use Outlier Detector for row-level scores on numeric columns. Choose Dataset Profiler for column types, counts, completeness, duplicates, statistics, date ranges, top values, and data-quality warnings. Choose Correlation Matrix Generator for pairwise Pearson, Spearman, or Kendall correlations. For ordered time-series data with timestamps, gaps, and rolling or seasonal thresholds, use Time-Series Anomaly Detector. These tools are part of Dataset Analyze Tools.

Frequently asked questions