Minimum Redundancy Maximum Relevance (MRMR) Algorithm for feature selection of regression models

Why does the Matlab's page on feature selection mention that MRMR method can be used only for classification models, when Matlab offers fsrmrmr function to use MRMR for feature selection of regression models?

Answer 1

[Drew]

In short,

• For classification, fscmrmr was introduced in R2019b. https://www.mathworks.com/help/stats/fscmrmr.html
• For regression, fsrmrmr was introduced in R2022a. https://www.mathworks.com/help/stats/fsrmrmr.html

So, for R2022a and later, it looks like the page https://www.mathworks.com/discovery/feature-selection.html needs to be updated to remove the "Only for classification" designation for MRMR.

See the "More About" and "Algorithms" sections of the fscmrmr and fsrmrmr doc pages for more details about the algorithms.