Hi @ClaudioSalvatoreArcidiacono

This functionality is already supported by sklearn through the f_regression.

I admit that f_regression is not great because it selects based on p_values and f_value instead of correlation coefficient, although the correlation coefficient could be inferred from the pvalue with the number of samples, so in essence, it is more or less the same. Highlight on more or less.

Hey @solegalli, thanks for taking a look and thanks for mentioning f_regression from sklearn, it is one of those functionalities that I never fully explored to be completely honest.

I believe that adding correlation with target as a selection method to SmartCorrelatedSelection could still bring value to the library for the following reasons:

• f_regression selects features to remove by looking at the interactions with the feature and the target, without considering correlations between features. In the proposed change the candidates features to drop are selected by looking at the correlation with other features and the correlation with target is used to make the final decision. So they achieve different purposes:
  • f_regression: Is used to remove features with a low correlation with the target.
  • SmartCorrelatedSelection with selection_method=corr_with_target as a selection method: Removes features with lower correlation with target among a correlated features group.

I would say that it is more like sklearn VarianceThreshold vs SmartCorrelatedSelection with variance as a selector method, they are quite different.

• Unlike f_regression, which is not designed as an estimator, SmartCorrelatedSelection is. This makes it easier for users to incorporate it directly into their feature selection pipelines.
• A direct correlation-based selection method can be more intuitive for many users, allowing them to easily understand and interpret the relationship between features and the target variable. This could enhance the usability and accessibility of the library.

Again, thanks for your time and I am happy to discuss this further!