Removing outliers from a matrix

I removed outliers from my each column of my feature matrix, but now I have feature vectors with differing lengths. I cannot make them the same length because I will lose data from a certain range. If I shuffle, then the correlation between my features and my outcome becomes messy. How can I do that? My matrix of features and outcomes is a 55000*14 matrix. The last column is the outputs.

[Davide Masiello]

Maybe, instead of removing the outliers, you could replace them with an interpolated value.

Answer 1

[Bjorn Gustavsson]

QD-answer: The best you can do for single pairs of columns is to only use the rows where neither are outliers. With the cov-function you can get this handled with the options 'omitrows' or 'partialrows' if you replace each outlier with a nan. If you want to use the corrcoef-function it has a slightly different interface. It takes a parameter-value pair to set the options for 'rows', see the help and documentation for description.

HTH