Outlier removal from a matrix

I removed the outliers from my dataset with rmoutliers(A,'mean') command. It should remove the data 3 standard deviations from the mean of each column. But when I print the histogram of each column, there are still some data as far as 6 standard deviations away. What do you suggest? Here is my code:

A = rmoutliers(table_data,'mean');

Zscores = zscore(A); %(A is a 50000*12 matrix)

figure

histogram(Zscores(:,2))

In the histogram, there are still some data as far as 6 standard deviations away.

[John D'Errico]

help rmoutliers

 RMOUTLIERS   Remove outliers from data
 
    B = RMOUTLIERS(A) detects and removes outliers from data. A can be a
    vector, matrix, table, or timetable. If A is a vector, RMOUTLIERS
    removes the entries detected as outliers. If A is a matrix or a table,
    RMOUTLIERS detects outliers for each column and then removes the rows
    containing outliers.
 
    B = RMOUTLIERS(A,METHOD) specifies the method used to determine
    outliers. METHOD must be one of the following: 'median' (default),
    'mean', 'quartiles', 'grubbs', or 'gesd'.
 
    B = RMOUTLIERS(A,'percentiles',[LP UP]) detects as outliers all
    elements which are below the lower percentile LP and above the upper
    percentile UP. LP and UP must be scalars between 0 and 100 with
    LP <= UP.
 
    B = RMOUTLIERS(A,MOVMETHOD,WL) uses a moving window method to determine
    contextual outliers instead of global outliers. MOVMETHOD can be
    'movmedian' or 'movmean'.
 
    B = RMOUTLIERS(...,DIM) reduces the size of A along the dimension DIM.
    Use DIM = 1 to remove rows and DIM = 2 to remove columns.
    RMOUTLIERS(A,DIM) first calls ISOUTLIER(A) to detect outliers.
 
    B = RMOUTLIERS(...,'MinNumOutliers',N) removes rows (columns) that
    contain at least N outliers. N must be an integer. By default, N = 1.
 
    B = RMOUTLIERS(...,'ThresholdFactor',P) modifies the outlier detection
    thresholds by a factor P. See the documentation for more information.
 
    B = RMOUTLIERS(...,'SamplePoints',X) specifies the sample points X
    representing the location of the data in A for the moving window
    methods 'movmedian' and 'movmean'. If the first input A is a table, X
    can also specify a table variable in A.
 
    B = RMOUTLIERS(...,'MaxNumOutliers',MAXN) specifies the maximum number
    of outliers for the 'gesd' method only.
 
    B = RMOUTLIERS(...,'OutlierLocations',INDOUTLIER) specifies the outlier
    locations according to the logical array INDOUTLIER. Elements of
    INDOUTLIER that are true indicate outliers in the corresponding element
    of A. INDOUTLIER must have the same size as A.
 
    [B,INDRM,INDOUTLIER,LTHRESH,UTHRESH,CENTER] = RMOUTLIERS(...) also
    returns a logical column (row) vector INDRM indicating which rows
    (columns) of A were removed, a logical array INDOUTLIER indicating the
    location of the detected outliers, and the lower threshold, upper
    threshold, and center value used by the outlier detection method.
 
    Arguments supported only for table inputs:
 
    B = RMOUTLIERS(...,'DataVariables',DV) removes rows (variables)
    according to outliers in table variables DV. The default is all table
    variables in A. DV must be a table variable name, a cell array of table
    variable names, a vector of table variable indices, a logical vector, a
    function handle that returns a logical scalar (such as @isnumeric), or
    a table vartype subscript.
 
    Examples:
 
      % Remove outliers from a vector
        a = [1 2 1000 3 4 5]
        b = rmoutliers(a)
 
      % Remove only the rows which contain at least 2 outliers
        A = [[1 2 1000 3 4 5]', [1 2 1000 3 4 1000]']
        [B,removedRows,outlierLocs] = rmoutliers(A,'MinNumOutliers',2)
 
    See also ISOUTLIER, FILLOUTLIERS, RMMISSING, ISMISSING, FILLMISSING

    Documentation for rmoutliers
       doc rmoutliers

    Other uses of rmoutliers

       gpuArray/rmoutliers    tall/rmoutliers

rmoutliers

I had to go to the doc to check your claim that rmoutliers with the 'mean' option does specifically use 3 standard deviations as the cutoff, away from the mean and then it removes the entire row containing that outlier. This is true. But rmoutliers is not a perfect tool, and any such tool can have problems if you dare to push its limits.

x = [ones(1,5),1 + eps,10]

x = 1×7

1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 10.0000

xhat = rmoutliers(x)

xhat = 1×5

1 1 1 1 1

xhat == 1

ans = 1×5 logical array
   1   1   1   1   1

So rmoutliers first removed the 10 as being more than 3 sigma out, but then, since the standard deviation of the first 5 elements is exactly zero, 1+eps is ALSO more than 3 sigma out, and a clear outlier. The point is, if you try hard enough, you can always cause any such adaptive tool to exhibit strange behavior.

But if you want to know what happened, then you need to provide your data. Otherwise, anything is just a wild guess.

Attach it to a comment (not as an answer), in a .mat file.