Matlab中二维数组的异常值分析问题求助

Hey there! Let's break down how to tackle this outlier filtering task in MATLAB step by step. It sounds like you’ve already calculated the mean and standard deviation for each column of your 2D array (since you have 5 values for each metric, I’m assuming your data has 5 columns—if it’s rows instead, we can tweak the code easily!).

Core Logic #

The requirement is to remove values that deviate from the mean by more than 1 standard deviation. For each column (or row) in your array:

• Calculate the lower threshold: mean - 1*std
• Calculate the upper threshold: mean + 1*std
• Any value outside this [lower, upper] range is considered an outlier and should be filtered out.

Step-by-Step MATLAB Code #

Let’s use your provided mean and std values in a concrete example:

% Replace this with your actual 2D array
data = ...; 

% Your precomputed mean and standard deviation values
col_means = [9.9900 10.0057 10.0180 10.0051 9.9312];
col_stds = [3.0257 3.0208 3.0119 2.9924 2.9868];

% Calculate threshold bounds for each column
lower_bounds = col_means - col_stds;
upper_bounds = col_means + col_stds;

% Create a logical mask where TRUE = outlier
outlier_mask = (data < lower_bounds) | (data > upper_bounds);

% Option 1: Replace outliers with NaN (preserves array structure)
data_cleaned = data;
data_cleaned(outlier_mask) = NaN;

% Option 2: Delete rows containing any outliers (adjust if filtering columns)
% data_cleaned = data(~any(outlier_mask, 2), :);

Key Explanations #

• Broadcasting: MATLAB automatically matches the column-wise thresholds to your 2D array, so you don’t need to loop through each column manually.
• Outlier Mask: The outlier_mask matrix makes it easy to identify and manipulate outliers without altering the original data until you’re ready.
• NaN vs. Row Deletion: Using NaN keeps your array’s shape intact, and you can use MATLAB’s nanmean, nanstd, etc., to ignore these values in later calculations. If you need to remove entire rows/columns with outliers, use the any() function to flag those rows.

Adjustment for Row-Wise Metrics #

If your mean and std values were calculated per row instead of per column, just transpose the threshold vectors to match the row dimensions:

lower_bounds = col_means' - col_stds';
upper_bounds = col_means' + col_stds';
outlier_mask = (data < lower_bounds) | (data > upper_bounds);

内容的提问来源于stack exchange，提问作者Amith Adiraju