Skip to main content

Feature Scaling Techniques

==Important to note that we should not be leaking the testing data into scaling. We won't know the test data in practice so they shouldn't be used to inform us.==

Standard scaler

  • Removing the mean and dividing by the standard deviation
  • This changes the feature to be centered around 0 and has unit variance
  • Many ML models assumes this to be true for the features Min Max Scale (Normalization)
  • Removing the minimum and dividing by the range
  • This changes the feature to be between 0 and 1 Normalizer
  • Scales each data point such that the feature vector has a Euclidean distance of 1
  • Used when the direction of the data matters but not the length of the vector #data-science #machine-learning