Machine Learning Overview (page 12 of 13)

Data Scaling refers to scale of our features. Certain machine learning algorithms might only work well if all features have the same scale. For instance, a distance algorithm such as k-nearest neighbors will not work well if one feature has a scale from 0 to 100K and another feature has a scale from 0 to 1. A few of the common scaling options are the following:

Standardization: scaled to have a mean of zero & stdev of 1
Normalization: scaled to have a unit mean and variance
Range: scaled to span a defined range
Binarization: above a threshold scaled to True, otherwise scaled to False