Theoretical Error in Feature Selection

[Subhasis-Biswas]

Feature selection is being done prior to train-test split, which can cause data leakage.

Link to discussion: https://stackoverflow.com/questions/56308116/should-feature-selection-be-done-before-train-test-split-or-after

--
Even though here only the correlation of features are being considered during feature selection (not regarding the target into consideration), the computation of correlation-coefficient itself requires the knowledge of the mean of the entire column, thus unknowingly (in this case) causing a leakage of test information to the train. This needs correction in my opinion.