Data Normalization

Typically, normalization refers to

  • transforming all values of some continuous variable to the same scale
  • it's done at the Data Transformation stage

There are several approaches

  • Min-Max
  • $Z$-score

Min-Max Normalization

Min-max normalization

  • normalize to scale $[\text{new_min}_A, \text{new_max}_A]$
  • for each new value, calculate $v'= \cfrac{v - \text{min}_A}{\text{max}_A - \text{min}_A} \cdot (\text{new_max}_A - \text{new_min}_A) + \text{new_min}_A$
  • the easiest model
  • not always good - if there are outliers


  • income range between 12K to 98K
  • want to normalize to $[0.0, 1.0]$.
  • so, for 73K have $\cfrac{73-12}{98-12} \approx 0.716$

$Z$-score Normalization

$v'= \cfrac{v - \mu_A}{\sigma_A}$


  • Assume that $\mu = 54K$ and $\sigma = 16K$
  • So 73K becomes 1.225



Machine Learning Bookcamp: Learn machine learning by doing projects. Get 40% off with code "grigorevpc".

Share your opinion