To combat potential collinearity issues, eliminate variables from the regression one at a time

[EgorKraevTransferwise]

Context

If we filter by pure statistical significance, we might end up, for example, dropping two features that are identical but highly predictive; this way, we eliminate one worst-t-value feature at a time from the regression so this won't happen.

Checklist

• [x ] Change meets or does not compromise the Baseline Security Requirements