Principal Component Analysis on a Dataset

Question

I have a dataset with 20 feature columns, 1 label column, and 1200 samples. I have to test an MLP classifier on this dataset. Does it make sense to apply PCA before using the MLP, given the fact that all the feature columns have a similar distribution, centered at 0 and with similar variances? My reasoning is that since PCA wants to reduce the dimensionality of the dataset by assessing which directions contain the most information (i.e. the most variance) it would be useless given that all columns have similar variances... Am I right?

Answer 1

The columns are not the only possible directions that PCA will "search" over. There are also linear combinations of columns that might explain a large portion of the variance compared to any single column.