Normalization is a way of stripping the units of the numerical scales. For example, if we have a feature vector  with height in meter as [10,20] and salary [5000,1000]

y value.JPG

In data mining we would want to scale both features to same same scale,  since we do not want the salary to influence more on the models.

Problem: How do we normalize ?
Answer : We can normalize in two ways.

Norm to one : where each vectors are normalized such that they sum to one i.e

height = [10 / sq.rt (10^2 + 20 ^2) …..]

norm to one.JPG

Normalizing using Z-score :  i.e  centering with standard deviation

Z-Score Normalisation.JPGZ-Score Normalisation eg.JPG

REF: https://www.utdallas.edu/~herve/abdi-Normalizing2010-pretty.pdf

 

 

 

Advertisements