Scaling

Many machine learning methods (SVM, neural networks, clustering algorithms, PCA) require data to be scaled. The values need to be brought into the same range, e.g. to make columns of a different order of magnitude comparable.

Two types of scaling are widely used:

Scaling to mean=0 and variance=1

The default of the scale function:

x <- c(1,2,3,4,5,6,7)

xscaled <- scale(x)

Scaling to mininum=0 and maximum=1

xscaled = scale(x, center=min(x), scale=max(x)-min(x))