Before you begin performing analytics on a dataset, it is important to identify and recognize outlier data patterns and values. 

Unusual values or patterns in the data can be sources for the following:

This section provides guidance in how to locate these patterns of data in individual columns.

Single-column outliers

For assessing anomalies in individual columns,  provides visual features and statistical information to quickly locate them.

Data Histogram

You can use the data quality bar and histogram to locate unusual values in your column data. The following example illustrates a dataset that contains two columns with outlier data. The first two rows are outliers with the subsequent rows to be consistently patterned data:

Numeric and string anomalies

Numeric data

The col-numbers column contains 100 random values 0-10, and singleton values -100 and 100.

In the histogram, you can see the outliers at the extremes of the graph. Note the slight visual distinction between the two extreme values and the values next to them, which are not represented in the column data.

Tip: In a histogram for numeric data, the spread between the extreme values and the more frequent values is a visual cue for outliers.

For numeric data, the range of values is displayed as part of the histogram. In this dataset, the extreme values are singletons. If a dataset contains more instances of outlier values, you should investigate further.

NOTE: In numeric datasets, a high count of outlier values may be statistically significant. You should review those values and related data in other columns before you perform operations to change or remove those rows.

Significant counts of unusual values

When your data contains a significant number of specific values, you should review them to see if the values have meaning. They may be placeholders for missing values. 

For numeric data, you should be skeptical of occurrences of the following values:

Suspicious valueReason
-1In system generated data, -1 is often an indicator of a failed result of some kind.
0Some systems will fill missing numeric values with the number 0. You should verify the meaning of the value of 0 in your dataset.
555-####In the United States, the phone number prefix 555 never corresponds to a person's phone number. These informational phone numbers and should not be considered as valid values for individuals' data.

In older versions of Microsoft Excel, 65,535 was the maximum number of rows permitted in a single sheet.

NOTE: 65,536 is 216, which is the maximum number of data bits in a 16-bit system.

2147483647This value is the largest positive integer that can be stored in an int datatype by 32-bit systems, which are still sources of data. If you see these values, the source system may have been unable to represent the true value and wrote this value instead.
4294967295This value is the largest raw value that can be stored in 32-bit systems. If you see these values, the source system may have been unable to represent the true value and wrote this value instead.
January 1st, 1900This value is the earliest date recognized by Microsoft Excel. The true date may not be accurately represented in your data.
January 1st, 1904
This value is the earliest date recognized by Microsoft Excel for Macintosh.
00:00:00 UTC on January 1, 1970This value is the earliest recognized date in UTC timestamp values. UTC timestamps are recorded as the number of milliseconds since this moment in time, stored as a signed 32-bit integer. Since datetime values may be represented in many different formats, you should identify these values for the date formats in your dataset.
03:14:07 UTC on Tuesday, 19 January 2038

This value is the latest recognized date in UTC timestamp values. Since datetime values may be represented in many different formats, you should identify these values for the date formats in your dataset.

  • This limit is generally known as the "Year 2038" problem.


String data

The col-strings column contains approximately 25 values for orange, red, greenyellow, and two instances of supercalifragilisticexpialidocious.

NOTE: For string-based data, outliers can be identified as strings with a low count of instances. These are the shorter stacks in the histogram.

Column Details

In the Column Details panel, you can review detailed statistics on the values in the currently selected column, including data on outliers. In the Transformer page, select Column Details from a column's drop-down.

Tip: In the Column Details panel, you can select specific outlier values, prompting suggestions, which enables you to take action on values identified by the platform as outliers.

Outliers in the Column Details

Column Detail Statistics

The Column Details panel provides information on the following:

NOTE: For string-based data types, these statistics pertain to string length.

Tip: Any green bar in the Column Details panel can be selected to prompt for suggestions on actions, including values in Outliers, Value Histogram, and Frequent Values graphs. Multi-select values as needed.


 uses a special set of computations to identify values that it designates as outliers. 

For more information on these computations and other calculations in the Column Details panel, see Column Statistics Reference.

Tune standard deviation calculations

Although standard deviation information is available in the Column Details, you may want to generate your own standard deviation calculation. For example, the following transform generates a new column which computes the number of standard deviations that a column value is from the average value for the column: 

You can then compute your own outlier function, using something like the following, which assumes that the above derived column has been renamed col_numbers_stdev and identifies outliers greater than 4 standard deviations from the average:

The above function generates boolean values in a new column, setting the value to true if the absolute value of the standard deviation for the col_numbers_stdev is more than 4. You can then perform operations based on the values written to this column or leave the column in place for downstream analytics tools.

The variance function is also supported.

Methods for fixing single-column outliers

After you have identified the values that are outliers in your column, you must determine if those values are valid or invalid for your dataset. For example, a value of 0 may be a valid measurement, or it may be a value that was inserted for lack of a valid value. 

For invalid values:

For valid values: