Using mathematical transformations

We sometimes want to use features that do not have a Gaussian distribution with a machine learning algorithm that assumes our features are distributed in that way. When that happens, we either need to change our minds about which algorithm to use (choose KNN or random forest rather than linear regression, for example) or transform our features so that they approximate a Gaussian distribution. We go over a couple of strategies for doing the latter in this recipe.

Getting ready

We will use the transformation module from feature engine in this recipe. We continue to work with the COVID-19 data, which has one row for each country with the total cases and deaths and some demographic data.

How to do it...

1. We start by importing the transformation module from feature_engine, train_test_split from sklearn, and stats from scipy. We also create a training and testing DataFrame with the COVID-19 data:

   import pandas as pd
   from feature_engine...