Data Discretization
Data discretization is a preprocessing technique used to transform continuous
data into discrete categories or bins.
1. Equal-Width Binning
2. Equal-Frequency Binning
3. Clustering-Based Binning
4. Decision Tree-Based Binning
5. Custom Binning
6. Entropy-Based Binning
7. Frequency-Based Binning
Data discretization
Data discretization is a useful technique for transforming continuous variables into categorical ones.
Here's a brief overview of each method:
► Equal-Width Binning: This method divides the range of the continuous variable into intervals of equal
width. For example, if you have data ranging from 0 to 100 and you want 5 bins, each bin would
cover a range of 20 units.
► Equal-Frequency Binning: This method divides the data into bins such that each bin contains
approximately the same number of data points. This is useful when you want each bin to have a
similar number of observations, making it more balanced.
► Clustering-Based Binning: This approach uses clustering algorithms (like k-means) to group data into
clusters, which are then used as bins. Each cluster represents a bin, and the data points within a
cluster are assigned to the same bin.
► Decision Tree-Based Binning: Decision trees can be used to create bins based on how well different
ranges of the data predict the target variable. The tree’s splits determine the bin edges, aiming to
maximize the separation between the bins in terms of the target variable.
► Custom Binning: This method involves defining bins based on domain knowledge or specific criteria.
The bins are manually set according to the problem's context or requirements.
► Entropy-Based Binning: This method involves creating bins that maximize the information gain or
reduce entropy. It’s often used in conjunction with decision trees and aims to produce bins that are
more informative with respect to the target variable.
► Frequency-Based Binning: Similar to equal-frequency binning, this method involves creating bins
based on the frequency of data points. However, it can be adjusted to group data points according to
specific frequency thresholds.
Each method has its strengths and is suited to different types of data and analytical goals.
Purpose of Binning
Binning or discretization serves several purposes in data preprocessing and analysis:
• Simplification: By converting continuous data into discrete bins, you simplify the dataset. This can make the
data easier to handle, visualize, and interpret, especially for non-technical stakeholders.
• Handling Outliers: Discretization can help manage outliers by grouping extreme values into the same bin,
thereby reducing their impact on the analysis.
• Improving Model Performance: In some machine learning algorithms, discrete features can perform better than
continuous features. For example, decision trees often work better with categorical variables.
• Data Compression: Binning can reduce the size of the data by summarizing continuous data into fewer discrete
categories. This can help in reducing storage and computational costs.
• Enhancing Interpretability: Discrete categories are often easier to interpret than continuous variables. Binning
can make it easier to understand patterns and relationships in the data.
• Reducing Noise: Discretization can smooth out noise in continuous data, making underlying patterns more
apparent.
• Facilitating Certain Analyses: Some statistical and machine learning techniques require discrete inputs or work
better with categorical data. Binning can transform continuous data to meet these requirements.
Overall, binning helps in making data analysis more manageable and can improve the performance of various
analytical methods.
Example Dataset
Let's say our dataset has the following house prices (in thousands of dollars):
[ 150, 200, 250, 300, 350, 400, 450, 500, 550, 600 ]
Goal: We want to discretize these prices into bins with equal width. Let’s choose to
create 3 bins.
Steps for Equal-Width Binning
• Determine the Range of Data:
• Calculate Bin Width:
• Define Bin Edges:
• Assign Data to Bins:
• Categorize Each Price:
[Link] the Range of Data:
Find the minimum and maximum values in the dataset.
• Minimum price = 150
• Maximum price = 600
Range = Maximum - Minimum = 600 - 150 = 450
[Link] Bin Width:
Number of bins = 3
Bin width = Range / Number of bins = 450 / 3 = 150
[Link] Bin Edges:
The bins will be defined by the edges calculated using the bin width.
• Bin 1: 150 to 150 + 150 = 150 to 300
• Bin 2: 300 to 300 + 150 = 300 to 450
• Bin 3: 450 to 450 + 150 = 450 to 600
[Link] Data to Bins:
Prices falling between 150 and 300 will be assigned to Bin 1.
Prices falling between 300 and 450 will be assigned to Bin 2.
Prices falling between 450 and 600 will be assigned to Bin 3
[Link] Each Price:
150 → Bin 1
200 → Bin 1
250 → Bin 1
300 → Bin 2
350 → Bin 2
400 → Bin 2
450 → Bin 3
500 → Bin 3
550 → Bin 3
600 → Bin 3
Output
► After applying equal-width binning, the house prices are categorized into the
following bins:
► -Bin 1 (150 to 300): 150, 200, 250
► -Bin 2 (300 to 450): 300, 350, 400
► -Bin 3 (450 to 600): 450, 500, 550, 600
► This process simplifies the continuous price data into discrete categories,
making it easier to analyze price ranges and identify trends in the house
e-commerce dataset.
Python code
import numpy as np
import pandas as pd
# Example dataset of house prices (in thousands of dollars)
house_prices = [Link]([150, 200, 250, 300, 350, 400, 450, 500, 550, 600])
# Define the number of bins
num_bins = 3
# Calculate the range and bin width
min_price = [Link](house_prices)
max_price = [Link](house_prices)
price_range = max_price - min_price
bin_width = price_range / num_bins
# Define the bin edges
bins = [Link](min_price, max_price + bin_width, bin_width)
# Categorize each house price into bins
bin_labels = [f'Bin {i+1}' for i in range(num_bins)]
binned_prices = [Link](house_prices, bins=bins, labels=bin_labels, include_lowest=True)
# Create a DataFrame to show the results
results = [Link]({
'House Price ($1000)': house_prices,
'Bin Number': binned_prices
})
# Display the results
print(results)