Machine Learning - Machine Learning Development Life Cycle - Binning and Binarization | Discretization | Quantile Binning | KMeans Binning Tutorial

1] Encoding Numerical Features

• Discretization (Binning)

Discretization is the process of transforming contiguous variable into discrete variables by creating a set of contiguous intervals that span the range of the variable’s values. Discretization is called binning, where bin is an alternative name for interval.

E.g suppose age is  12,13,13,14,24,28,29,30,30,32,32,26,45.

Then discretization is a process of creating bin i.e 10-20,20-30,30-40,40-50, etc

Why use Discretization:

1] To Handle Outlier

2] To improve the value spread

Class- Sklearn - KbinsDiscretizer

Types Of Discretization-

1] Unsupervised

• Equal Width (uniform)

E.g suppose age is  12,13,13,14,24,28,29,30,30,32,32,26,45.

Suppose you need to create 5 bins, max value is 50, min value is 10

Then (max-min)/bins = 50-10/5 = 8

Then your each interval will be of uniform length 8

i.e 10-18, 18-26, 26-34, 34-42, 42-50  ----Total 5 equal width bin

therefore the age will be-

10-18 – 4

18-26 – 1

26-34 – 7

34-42 – 0

42-50 – 1

• Equal Frequency (quantile)

• K means 

This will create clusters

2] Supervised

• Decision Tree

3] Custom Binning

It is binning using custom range

[0 – 18] – Kids

[18 – 60] – Adult

[60 - 80 ] – Senior  

• Binarization

Converting continuous value to binary

Class – Scikit Learn  - Binarizer

Eg  salary<6L – 1

      Salary>6L – 0