Standardization VS Normalization
Difference between Standardization and Normalization , formula, when to use.
Introduction:
Standardization and Normalization both are feature-scaling techniques that are used to convert the data into a shortened range to achieve faster convergence and improve accuracy of the distance based algorithms such as KNN, SVM etc. as the features may a huge numerical difference between them.
For example: The age of a person and their salary will have a significant numerical difference which may impact the performance of our model. To resolve this issue, we can perform feature scaling.
Standardization:
Standardization is a Linear Transformation technique, that allows us to transform the data in a uniform manner by making the:
Mean of the feature = 0 and Standard deviation = 1
It is also known as Z-Score
Standardization can be implemented by using the sklearn library as: from sklearn.preprocessing import StandardScaler
When to use Standardization:
We can use Standardization when our data follows Normal or Gaussian Distribution. In other words, the graph of the features makes a bell shaped curve.
One thing to remember here is that Standardization DOES NOT makes a distribution Normal or Gaussian Distribution.
Normalization:
Normalization is a scaling technique that is used to bring all the data points in the range of 0 to 1.
Normalization can be implemented by using the sklearn library as: from sklearn.preprocessing import MinMaxScaler
When to use Normalization :
Normalization can be used when our data does not follows Gaussian or Normal Distribution instead the graph makes a shape as:
One thing to remember here is that normalization is heavily influenced by the outliers in the data. Therefore the data should be treated for outliers before performing normalization.