I have a question about simulating data with multivariate normal distribution. I want to generate data for a binary classification problem. Can I generate a subset A with parameters mi_a and sigma_a and assign it class 1, next generate a subset B with parameters mi_b and sigma_b=sigma_a and assign it class 2. And then combine these subsets into one set?

It's okay? Do I have to do it in a different way, such as using marginal distributions?

  • $\begingroup$ A multivariate normal distribution will often involve non-zero covariances $\endgroup$
    – Henry
    May 9, 2021 at 14:25
  • $\begingroup$ That's fine. You can create toy datasets however you want. Out of curiosity, what's your purpose for generating this data? $\endgroup$
    – littleO
    May 9, 2021 at 14:27
  • $\begingroup$ Do you have a covariance matrix? If you don't, what you do is fine. Otherwise you have to use something like Cholesky or spectral decomposition, to apply the proper correlations. $\endgroup$
    – Momo
    May 9, 2021 at 14:31
  • $\begingroup$ I analyze different methods of classification of unbalanced data. And I want to compare these methods on real and synthetic data to consider different scenarios (class separation, noise, size etc). $\endgroup$
    – Marni
    May 9, 2021 at 14:32
  • $\begingroup$ Thank you for your answers :) $\endgroup$
    – Marni
    May 9, 2021 at 14:36


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.