A common problem when developing ML models is deciding which features should be used when training a model. For a supervised learning algorithm, the best features are those that are highly correlated with the label variable. This means, broadly speaking, that changing one of the variables induces a change in the other variable as well. An example of highly correlated variables could be the time of day and the amount of road traffic: traffic jams usually occur during the rush hour, while the amount of traffic during the night is particularly low.
The general aim of feature selection is to discover the variables that have the largest impact on the target variable. If the input dataset contains a large amount of columns, it ...
