isoutlier

Find outliers in data

Syntax

TF = isoutlier(A)

TF = isoutlier(A,method)

TF = isoutlier(A,'percentiles',threshold)

TF = isoutlier(A,movmethod,window)

TF = isoutlier(___,dim)

TF = isoutlier(___,Name,Value)

[TF,L,U,C] = isoutlier(___)

Description

TF = isoutlier(A) returns a logical array whose elements are true when an outlier is detected in the corresponding element of A. By default, an outlier is a value that is more than three scaled median absolute deviations (MAD) away from the median. If A is a matrix or table, then isoutlier operates on each column separately. If A is a multidimensional array, then isoutlier operates along the first dimension whose size does not equal 1.

TF = isoutlier(A,method) specifies a method for detecting outliers. For example, isoutlier(A,'mean') returns true for all elements more than three standard deviations from the mean.

TF = isoutlier(A,'percentiles',threshold) defines outliers as points outside of the percentiles specified in threshold. The threshold argument is a two-element row vector containing the lower and upper percentile thresholds, such as [10 90].

TF = isoutlier(A,movmethod,window) specifies a moving method for detecting local outliers according to a window length defined by window. For example, isoutlier(A,'movmedian',5) returns true for all elements more than three local scaled MAD from the local median within a sliding window containing five elements.

TF = isoutlier(___,dim) operates along dimension dim of A for any of the previous syntaxes. For example, isoutlier(A,2) operates on each row of a matrix A.

TF = isoutlier(___,Name,Value) specifies additional parameters for detecting outliers using one or more name-value pair arguments. For example, isoutlier(A,'SamplePoints',t) detects outliers in A relative to the corresponding elements of a time vector t.

[TF,L,U,C] = isoutlier(___) also returns the lower and upper thresholds and the center value used by the outlier detection method.

Examples

Detect Outliers in Vector

Find the outliers in a vector of data. A logical 1 in the output indicates the location of an outlier.

A = [57 59 60 100 59 58 57 58 300 61 62 60 62 58 57];
TF = isoutlier(A)

TF = 1x15 logical array

   0   0   0   1   0   0   0   0   1   0   0   0   0   0   0

Detect Outliers using Mean

Define outliers as points more than three standard deviations from the mean, and find the locations of outliers in a vector.

A = [57 59 60 100 59 58 57 58 300 61 62 60 62 58 57];
TF = isoutlier(A,'mean')

TF = 1x15 logical array

   0   0   0   0   0   0   0   0   1   0   0   0   0   0   0

Detect Outliers with Sliding Window

Create a vector of data containing a local outlier.

x = -2*pi:0.1:2*pi;
A = sin(x);
A(47) = 0;

Create a time vector that corresponds to the data in A.

t = datetime(2017,1,1,0,0,0) + hours(0:length(x)-1);

Define outliers as points more than three local scaled MAD away from the local median within a sliding window. Find the locations of the outliers in A relative to the points in t with a window size of 5 hours. Plot the data and detected outliers.

TF = isoutlier(A,'movmedian',hours(5),'SamplePoints',t);
plot(t,A,t(TF),A(TF),'x')
legend('Data','Outlier')

Matrix of Data

Find outliers for each row of a matrix.

Create a matrix of data containing outliers along the diagonal.

A = magic(5) + diag(200*ones(1,5))

A = 5×5

   217    24     1     8    15
    23   205     7    14    16
     4     6   213    20    22
    10    12    19   221     3
    11    18    25     2   209

Find the locations of outliers based on the data in each row.

TF = isoutlier(A,2)

TF = 5x5 logical array

   1   0   0   0   0
   0   1   0   0   0
   0   0   1   0   0
   0   0   0   1   0
   0   0   0   0   1

Compute Outlier Thresholds

Create a vector of data containing an outlier. Find and plot the location of the outlier, and the thresholds and center value determined by the outlier method. The center value is the median of the data, and the upper and lower thresholds are three scaled MAD above and below the median.

x = 1:10;
A = [60 59 49 49 58 100 61 57 48 58];
[TF,L,U,C] = isoutlier(A);
plot(x,A,x(TF),A(TF),'x',x,L*ones(1,10),x,U*ones(1,10),x,C*ones(1,10))
legend('Original Data','Outlier','Lower Threshold','Upper Threshold','Center Value')

Input Arguments

A — Input data

vector | matrix | multidimensional array | table | timetable

Input data, specified as a vector, matrix, multidimensional array, table, or timetable.

If A is a table, then its variables must be of type double or single, or you can use the 'DataVariables' name-value pair to list double or single variables explicitly. Specifying variables is useful when you are working with a table that contains variables with data types other than double or single.

If A is a timetable, then isoutlier operates only on the table elements. Row times must be unique and listed in ascending order.

Data Types: double | single | table | timetable

method — Method for detecting outliers

'median' (default) | 'mean' | 'quartiles' | 'grubbs' | 'gesd'

Method for detecting outliers, specified as one of the following:

Method	Description
'median'	Returns true for elements more than three scaled MAD from the median. The scaled MAD is defined as c*median(abs(A-median(A))), where c=-1/(sqrt(2)*erfcinv(3/2)).
'mean'	Returns true for elements more than three standard deviations from the mean. This method is faster but less robust than 'median'.
'quartiles'	Returns true for elements more than 1.5 interquartile ranges above the upper quartile or below the lower quartile. This method is useful when the data in A is not normally distributed.
'grubbs'	Applies Grubbs’s test for outliers, which removes one outlier per iteration based on hypothesis testing. This method assumes that the data in A is normally distributed.
'gesd'	Applies the generalized extreme Studentized deviate test for outliers. This iterative method is similar to 'grubbs', but can perform better when there are multiple outliers masking each other.

threshold — Percentile thresholds

two-element row vector

Percentile thresholds, specified as a two-element row vector whose elements are in the interval [0,100]. The first element indicates the lower percentile threshold and the second element indicates the upper percentile threshold. For example, a threshold of [10 90] defines outliers as points below the 10th percentile and above the 90th percentile. The first element of threshold must be less than the second element.

movmethod — Moving method

'movmedian' | 'movmean'

Moving method for detecting outliers, specified as one of the following:

Method	Description
'movmedian'	Returns true for elements more than three local scaled MAD from the local median over a window length specified by window. This method is also known as a Hampel filter.
'movmean'	Returns true for elements more than three local standard deviations from the local mean over a window length specified by window.

window — Window length

positive integer scalar | two-element vector of positive integers | positive duration scalar | two-element vector of positive durations

Window length, specified as a positive integer scalar, a two-element vector of positive integers, a positive duration scalar, or a two-element vector of positive durations.

When window is a positive integer scalar, the window is centered about the current element and contains window-1 neighboring elements. If window is even, then the window is centered about the current and previous elements.

When window is a two-element vector of positive integers [b f], the window contains the current element, b elements backward, and f elements forward.

When A is a timetable or 'SamplePoints' is specified as a datetime or duration vector, then window must be of type duration, and the windows are computed relative to the sample points.

Data Types: double | single | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64 | duration

dim — Dimension to operate along

positive integer scalar

Dimension to operate along, specified as a positive integer scalar. If no value is specified, then the default is the first array dimension whose size does not equal 1.

Consider a matrix A.

isoutlier(A,1) detects outliers based on the data in each column of A.

isoutlier(A,2) detects outliers based on the data in each row of A.

When A is a table or timetable, dim is not supported. isoutlier operates along each table or timetable variable separately.

Data Types: double | single | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64

Name-Value Pair Arguments

Specify optional comma-separated pairs of Name,Value arguments. Name is the argument name and Value is the corresponding value. Name must appear inside quotes. You can specify several name and value pair arguments in any order as Name1,Value1,...,NameN,ValueN.

Example: isoutlier(A,'mean','ThresholdFactor',4)

Data Options

'SamplePoints' — Sample points

vector

Sample points, specified as the comma-separated pair consisting of 'SamplePoints' and a vector. The sample points represent the x-axis locations of the data in A, and must be sorted and contain unique elements. Sample points do not need to be uniformly sampled. The default sample points vector is [1 2 3 ...].

Note

This name-value pair is not supported when the input data is a timetable. Timetables always use the vector of row times as the sample points. To use different sample points, you must edit the timetable so that the row times contain the desired sample points.

Moving windows are defined relative to the sample points. For example, if t is a vector of times corresponding to the input data, then isoutlier(rand(1,10),'movmean',3,'SamplePoints',t) has a window that represents the time interval between t(i)-1.5 and t(i)+1.5.

When the sample points vector has data type datetime or duration, then the moving window length must have type duration.

Example: isoutlier(T,'SamplePoints',0:0.1:10)

Data Types: single | double | datetime | duration

'DataVariables' — Table variables to operate on

scalar | vector | cell array | function handle | table vartype subscript

Table variables to operate on, specified as the comma-separated pair consisting of 'DataVariables' and one of the options in this table. The 'DataVariables' value indicates which variables of the input table to examine for outliers. The data type associated with the indicated variables must be double or single. Other variables in the table not specified by 'DataVariables' are not operated on, so the output contains false values for those variables.

Option	Description	Examples
Variable name	A character vector or scalar string specifying a single table variable name	'Var1' "Var1"
Vector of variable names	A cell array of character vectors or string array where each element is a table variable name	{'Var1' 'Var2'} ["Var1" "Var2"]
Scalar or vector of variable indices	A scalar or vector of table variable indices	1 [1 3 5]
Logical vector	A logical vector whose elements each correspond to a table variable, where true includes the corresponding variable and false excludes it	[true false true]
Function handle	A function handle that takes a table variable as input and returns a logical scalar	@isnumeric
vartype subscript	A table subscript generated by the vartype function	vartype('numeric')

Example: isoutlier(T,'DataVariables',["Var1" "Var2" "Var4"])

Outlier Detection Options

'ThresholdFactor' — Detection threshold factor

nonnegative scalar

Detection threshold factor, specified as the comma-separated pair consisting of 'ThresholdFactor' and a nonnegative scalar.

For methods 'median' and 'movmedian', the detection threshold factor replaces the number of scaled MAD, which is 3 by default.

For methods 'mean' and 'movmean', the detection threshold factor replaces the number of standard deviations from the mean, which is 3 by default.

For methods 'grubbs' and 'gesd', the detection threshold factor is a scalar ranging from 0 to 1. Values close to 0 result in a smaller number of outliers and values close to 1 result in a larger number of outliers. The default detection threshold factor is 0.05.

For the 'quartiles' method, the detection threshold factor replaces the number of interquartile ranges, which is 1.5 by default.

This name-value pair is not supported when the specified method is 'percentiles'.

Data Types: double | single | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64

'MaxNumOutliers' — Maximum outlier count

positive integer

Maximum outlier count, for the 'gesd' method only, specified as the comma-separated pair consisting of 'MaxNumOutliers' and a positive integer. The 'MaxNumOutliers' value specifies the maximum number of outliers returned by the 'gesd' method. For example, isoutlier(A,'gesd','MaxNumOutliers',5) returns no more than five outliers.

The default value for 'MaxNumOutliers' is the integer nearest to 10 percent of the number of elements in A. Setting a larger value for the maximum number of outliers can ensure that all outliers are detected, but at the cost of reduced computational efficiency.

The 'gesd' method assumes the non-outlier input data is sampled from an approximate normal distribution. When the data is not sampled in this way, the number of returned outliers might exceed the 'MaxNumOutliers' value.

Data Types: double | single | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64

Output Arguments

TF — Outlier indicator

vector | matrix | multidimensional array

Outlier indicator, returned as a vector, matrix, or multidimensional array. An element of TF is true when the corresponding element of A is an outlier and false otherwise. TF is the same size as A.

Data Types: logical

L — Lower threshold

scalar | vector | matrix | multidimensional array | table | timetable

Lower threshold used by the outlier detection method, returned as a scalar, vector, matrix, multidimensional array, table, or timetable. For example, the lower value of the default outlier detection method is three scaled MAD below the median of the input data. L has the same size as A in all dimensions except for the operating dimension where the length is 1.

Data Types: double | single | table | timetable

U — Upper threshold

scalar | vector | matrix | multidimensional array | table | timetable

Upper threshold used by the outlier detection method, returned as a scalar, vector, matrix, multidimensional array, table, or timetable. For example, the upper value of the default outlier detection method is three scaled MAD above the median of the input data. U has the same size as A in all dimensions except for the operating dimension where the length is 1.

Data Types: double | single | table | timetable

C — Center value

scalar | vector | matrix | multidimensional array | table | timetable

Center value used by the outlier detection method, returned as a scalar, vector, matrix, multidimensional array, table, or timetable. For example, the center value of the default outlier detection method is the median of the input data. C has the same size as A in all dimensions except for the operating dimension where the length is 1.

Data Types: double | single | table | timetable

More About

Median Absolute Deviation

For a random variable vector A made up of N scalar observations, the median absolute deviation (MAD) is defined as

$$\text{MAD = median}\left(|A_i-\text{median}\left(A\right)|\right)$$

for i = 1,2,...,N.

The scaled MAD is defined as c*median(abs(A-median(A))) where c=-1/(sqrt(2)*erfcinv(3/2)).

Extended Capabilities

Tall Arrays

Calculate with arrays that have more rows than fit in memory.

Usage notes and limitations:

• The 'percentiles', 'grubbs', and 'gesd' methods are not supported.

• The 'movmedian' and 'movmean' methods do not support tall timetables.

• The 'SamplePoints' and 'MaxNumOutliers' name-value pairs are not supported.

• The value of 'DataVariables' cannot be a function handle.

• Computation of isoutlier(A), isoutlier(A,'median',...), or isoutlier(A,'quartiles',...) along the first dimension is only supported for tall column vectors A.

For more information, see Tall Arrays.

C/C++ Code Generation

Generate C and C++ code using MATLAB® Coder™.

Usage notes and limitations:

• The 'movmean' and 'movmedian' methods for detecting outliers do not support timetable input data, datetime 'SamplePoints' values, or duration 'SamplePoints' values.

• String and character array inputs must be constant.

GPU Arrays

Accelerate code by running on a graphics processing unit (GPU) using Parallel Computing Toolbox™.

Usage notes and limitations:

• The 'movmedian' moving method is not supported.

• The 'SamplePoints' and 'DataVariables' name-value pairs are not supported.

For more information, see Run MATLAB Functions on a GPU (Parallel Computing Toolbox).