Cluster Analysis in Data Mining: Applications, Methods & Requirements [With Examples]

As technology grows, so does the amount of data we create. According to Statista, global data creation is expected to exceed 180 zettabytes by 2025. With so much data, cluster analysis in data mining can help.

Cluster analysis organizes data by grouping similar items together. It helps us find patterns and sort information more easily. It is similar to putting things into different boxes based on their common characteristics.

Cluster analysis in data mining is used in many areas:

• Businesses use it to group customers with similar buying habits.
• Image processing tools use it to separate parts of an image.
• Social media platforms use it to recommend content to users.

In this blog, we’ll go through the basics of cluster analysis, look at how it’s used, and explain the main methods and tools. Real-life examples will show how clustering helps in organizing data and finding patterns that are easy to understand.

Check out: Common Examples of Data Mining.

What is a Cluster?

A cluster is a way to group similar things together and keep different things separate. Clusters help organize and categorize items based on what they have in common. For example, imagine sorting a mix of fruits and vegetables. You might group all the vegetables together in one cluster and all the fruits in another. In this case, fruits and vegetables are two clusters, each holding similar items. 

What is Clustering in Data Mining?

Clustering in data mining is a technique used to divide data into meaningful subgroups or “clusters.” This method helps organize large datasets by identifying patterns and similarities among data points, even without predefined labels.

Clustering is a type of unsupervised learning because it doesn’t require known labels or categories in advance. Instead, clustering finds patterns on its own by grouping similar data points together based on shared characteristics.
 

Properties of Clustering

Clustering is useful for finding patterns without needing prior knowledge of class labels. Good clustering relies on key properties that define how well the data is grouped. Here’s a breakdown of these essential properties:

• Homogeneity: Data points within each cluster are similar to each other. This similarity gives each cluster a unique set of shared characteristics.
• Separation: Different clusters are distinct from one another, with clear boundaries. Good separation keeps clusters from overlapping and makes it easier to analyze each group.
• Compactness: Data points in each cluster are close together. This forms tight, well-defined groups that are easy to interpret.
• Connectedness: Points within a cluster are more closely connected to each other than to points in other clusters. This creates strong relationships within each group.

Different clustering methods focus on these properties in different ways. For example:

• K-Means Clustering groups data points around a defined number of central points, which creates compact clusters.
• Hierarchical Clustering gradually builds clusters in a layered structure, making groups that reflect natural connections within the data.
• Density-Based Clustering focuses on dense areas of data and creates clusters that are compact and separated.

Importance of Cluster Analysis in Data Mining

Cluster analysis helps make sense of large sets of data by grouping similar data points together. This approach reveals patterns and insights that may not be obvious at first glance. Here’s why cluster analysis is valuable:

• Customer Segmentation: Around 90% of companies use clustering to understand their customers better. It helps them group customers by shopping habits, interests, or preferences. This way, businesses can create offers that match each group’s needs.
• Market Research: Clustering helps businesses spot specific groups in a market, like age groups or regional preferences. With this, they can design products or services that speak directly to each group.
• Image and Visual Recognition: In healthcare, clustering is used to highlight key areas in medical images, like identifying organs in a scan. It also helps in apps that recognize faces or objects, making image processing faster and more accurate.
• Detecting Unusual Patterns: Clustering helps find unusual data points. For example, banks use it to spot suspicious transactions that don’t fit typical patterns, helping them detect potential fraud.
• Simplifies Large Data Sets: Clustering organizes data into smaller, related groups. This makes it easier to analyze and keeps analysts from getting lost in too much detail.
• Social Media Recommendations: Social platforms use clustering to group users by common interests. This makes it easy to suggest friends, groups, or content that each user might enjoy.

Clustering Methods in Data Mining

Clustering methods in data mining are used to group data points into meaningful clusters. Each method has a different approach, suitable for various types of data and clustering goals. Here’s an overview of key clustering methods:

1. Partitioning Method

The Partitioning Method divides the dataset into non-overlapping clusters or subsets. Each data point belongs to only one cluster, and there are no empty clusters. This method works by iteratively moving data points between clusters to improve clustering quality.

Characteristics:

• Requires the number of clusters (k) to be specified in advance.
• Ensures each data point belongs to a single cluster.
• Uses iterative relocation to optimize clusters, meaning data points are moved between clusters to improve the quality of clustering.

Example Algorithm: K-Means Clustering

• K-Means divides data into k clusters by finding k centroids (cluster centers) and assigning each data point to the nearest centroid. This process continues until the centroids stabilize, creating well-defined clusters based on proximity to each centroid.

Python Code Example:

This example uses K-Means clustering to divide the data into two clusters.

from sklearn.cluster import KMeans
import numpy as np

# Sample data
data = np.array([[1, 2], [2, 3], [4, 5], [6, 8], [7, 9]])

# Applying K-means with 2 clusters
kmeans = KMeans(n_clusters=2, random_state=0).fit(data)

# Output cluster labels and centroids
print("Cluster Labels:", kmeans.labels_)
print("Centroids:", kmeans.cluster_centers_)

Output:

lua

Cluster Labels: [1 1 1 0 0]
Centroids: [[6.5 8.5]
            [2.33333333 3.33333333]]

In this example, data points are divided into two clusters, with the centroids located at approximately [6.5, 8.5] and [2.33, 3.33].

2. Hierarchical Method

The Hierarchical Method creates a tree-like structure (or dendrogram) of clusters by progressively merging or splitting clusters. This method does not require a predefined number of clusters, and the process continues until each data point is either in its own cluster or all points form a single cluster.

Types of Hierarchical Clustering:

• Agglomerative Approach (Bottom-Up): Starts with each data point as its own cluster, then repeatedly merges the closest clusters until a stopping criterion is reached.
• Divisive Approach (Top-Down): Begins with all data points in a single cluster and splits clusters recursively until each data point is isolated or a stopping criterion is met.

Advantages:

• No need to specify the number of clusters in advance.
• Suitable for datasets where hierarchical relationships between data points are of interest.

Python Code Example:

This example performs hierarchical clustering and generates a dendrogram to visualize the clustering structure.

from scipy.cluster.hierarchy import dendrogram, linkage
import matplotlib.pyplot as plt
import numpy as np

# Sample data
data = np.array([[1, 2], [2, 3], [4, 5], [6, 8], [7, 9]])

# Perform hierarchical clustering using 'ward' linkage method
Z = linkage(data, method='ward')

# Plot dendrogram
plt.figure(figsize=(8, 4))
dendrogram(Z)
plt.title('Hierarchical Clustering Dendrogram')
plt.xlabel('Data Points')
plt.ylabel('Distance')
plt.show()

Output (Dendrogram Plot): The dendrogram visually represents the hierarchical clustering process, showing how clusters are merged based on distance. The plot displays data points on the x-axis and the merging distance on the y-axis, with clusters joined at different heights.

3. Density-Based Method

The Density-Based Method identifies clusters based on the density of data points. In this method, clusters are formed where data points are closely packed together, separated by areas of lower data density. This method is effective for discovering clusters of varying shapes and identifying outliers.

Example Algorithm: DBSCAN (Density-Based Spatial Clustering of Applications with Noise)

• DBSCAN forms clusters based on two parameters: epsilon (the maximum distance between points in a cluster) and min_samples (the minimum number of points required to form a dense region).
• Points in dense regions are grouped as clusters, while points that don’t meet the density requirement are marked as noise (outliers).

Advantages:

• Does not require the number of clusters to be specified beforehand.
• Effective for datasets with noise and irregularly shaped clusters.

Python Code Example:

This example applies DBSCAN to detect clusters based on density, with a sample dataset that includes some noise.

from sklearn.cluster import DBSCAN
import numpy as np

# Sample data
data = np.array([[1, 2], [2, 3], [4, 5], [6, 8], [7, 9], [10, 10]])

# Applying DBSCAN
dbscan = DBSCAN(eps=2, min_samples=2).fit(data)

# Output cluster labels
print("Cluster Labels:", dbscan.labels_)

Output:

Cluster Labels: [0 0 0 1 1 -1]

In this example, DBSCAN forms two clusters (labeled 0 and 1). The point [10, 10] is labeled -1, indicating it is an outlier (noise) because it doesn’t meet the density requirements to join a cluster.

4. Grid-Based Method

The Grid-Based Method divides the data space into a grid structure, where each cell represents a specific area of the data space. Clustering is then performed on these cells rather than on individual data points. This method is especially useful for spatial data, as it organizes data points based on their spatial location.

Key Points:

• Grid Structure: Data space is quantized into a finite number of cells that form a grid.
• Faster Processing: Since clustering is based on cells rather than individual points, the process is faster and depends on the number of cells in the grid, not on the dataset size.

Advantages:

• Quick processing time, as it clusters based on the grid cell density.
• Suitable for handling large spatial datasets.

Python Code Example:

Here’s a basic example demonstrating how grid-based clustering might look with a simplified grid approach. This is not a built-in grid-based clustering function, but rather a way to represent a grid-based approach.

import numpy as np
import matplotlib.pyplot as plt

# Sample spatial data
data = np.array([[1, 2], [2, 3], [4, 5], [10, 10], [12, 14], [13, 13]])

# Defining grid parameters
x_bins = np.linspace(min(data[:,0]), max(data[:,0]), 4)  # Divide x-axis into 3 bins
y_bins = np.linspace(min(data[:,1]), max(data[:,1]), 4)  # Divide y-axis into 3 bins

# Assign each point to a grid cell
x_indices = np.digitize(data[:,0], x_bins) - 1
y_indices = np.digitize(data[:,1], y_bins) - 1

# Visualize grid clustering
plt.scatter(data[:,0], data[:,1], c="blue")
plt.vlines(x_bins, min(data[:,1]), max(data[:,1]), colors="gray", linestyles="dotted")
plt.hlines(y_bins, min(data[:,0]), max(data[:,0]), colors="gray", linestyles="dotted")
plt.title("Grid-Based Clustering Example")
plt.xlabel("X-Axis")
plt.ylabel("Y-Axis")
plt.show()

# Output each point's assigned grid cell
for i, (x_idx, y_idx) in enumerate(zip(x_indices, y_indices)):
    print(f"Point {data[i]} is in grid cell ({x_idx}, {y_idx})")

5. Model-Based Method

The Model-Based Method assumes that each cluster follows a statistical model, often using Gaussian Mixture Models (GMM). This method finds clusters based on the probability that data points belong to a specific model, such as a Gaussian distribution. It also adjusts for noise and can determine the optimal number of clusters based on statistical measures.

Key Points:

• Assumes a Model: Uses a predefined model (e.g., Gaussian) for each cluster.
• Flexibility with Noise: Can account for noise and identify outliers.
• Automatic Cluster Count: Determines the number of clusters based on model parameters and data distribution.

Python Code Example:

from sklearn.mixture import GaussianMixture
import numpy as np

# Sample data
data = np.array([[1, 2], [2, 3], [4, 5], [10, 10], [12, 14], [13, 13]])

# Applying Gaussian Mixture Model with 2 clusters
gmm = GaussianMixture(n_components=2, random_state=0).fit(data)

# Predict cluster labels
labels = gmm.predict(data)
print("Cluster Labels:", labels)
print("Means of Each Cluster:", gmm.means_)

Output:

lua

Cluster Labels: [1 1 1 0 0 0]
Means of Each Cluster: [[11.66666667 12.33333333]
                        [2.33333333 3.33333333]]

This example shows how Gaussian Mixture Models (GMM) cluster the data into two groups, with calculated cluster centers based on the mean of each group.

Must read: Learn excel online free!

6. Constraint-Based Method

The Constraint-Based Method incorporates user-specified constraints to guide the clustering process. Constraints can include minimum cluster size, distance thresholds, or other specific properties required for the clusters. This approach is often used in applications where domain knowledge is crucial for meaningful clustering.

Key Points:

• User Constraints: Clustering is based on user-defined requirements, such as specific distance functions or cluster sizes.
• Interactive Approach: Allows users to adjust clustering parameters based on their needs.
• Application-Specific: Useful for applications requiring precise control over clustering results.

Example Scenario (Conceptual Example with Custom Constraints): Here’s a simplified conceptual example to demonstrate constrained clustering. Note that there’s no built-in function for constraint-based clustering, but custom constraints can be applied.

python

import numpy as np
from sklearn.cluster import KMeans

# Sample data
data = np.array([[1, 2], [2, 3], [4, 5], [6, 7], [10, 12], [12, 14]])

# Custom constraint: Minimum cluster size of 3
min_cluster_size = 3

# Applying K-Means with 2 clusters, then filtering by custom constraint
kmeans = KMeans(n_clusters=2, random_state=0).fit(data)
labels = kmeans.labels_

# Check cluster sizes
cluster_counts = np.bincount(labels)
valid_clusters = [i for i, count in enumerate(cluster_counts) if count >= min_cluster_size]

# Output valid clusters based on the minimum size constraint
print("Cluster Labels:", labels)
print("Cluster Counts:", cluster_counts)
print("Valid Clusters (min size 3):", valid_clusters)

Output:

Cluster Labels: [1 1 1 0 0 0]
Cluster Counts: [3 3]
Valid Clusters (min size 3): [0, 1]

This example applies a basic constraint, filtering clusters to only include those meeting the minimum size. Custom constraints can be adapted based on application needs.

Business Applications of Cluster Analysis

Cluster analysis in data mining is widely used across industries, with some impressive results. Here’s how different sectors apply cluster analysis to achieve their goals:

Read more about the applications of data science in finance industry.

Why Validate Clusters?

Validation helps confirm that clusters provide real insights rather than random groupings. This step checks for quality, ensuring that clusters are well-formed and distinct from one another.

Types of Cluster Validation

1. Internal Validation
   • Purpose: Measures the quality of clusters without using external or labeled data. Internal validation helps determine how well-defined and distinct clusters are based solely on the dataset.
   • How It Works: Internal validation uses metrics like compactness, separation, and connectivity to evaluate clusters.
   • Limitation: Internal validation can show relative quality but does not confirm how accurately clusters reflect real categories, as it does not use external labels.
2. External Validation
   • Purpose: Compares clusters to a known labeled dataset to check for accuracy.
   • How It Works: Applies clustering to a labeled dataset and compares the clusters formed to the actual categories. This method helps determine how well the clustering reflects true groupings in the data.
   • Limitation: External validation relies on labeled data, which isn’t always available and might favor certain clustering methods over others.

Key Metrics for Cluster Validation

Example of Clustering Algorithm

In this example, we’ll implement K-Means Clustering from start to finish using Python. This walk-through will explain each step, from preparing the data to interpreting the output.

• Step 1: Import Libraries

First, we’ll import the necessary libraries: numpy for handling arrays and KMeans from sklearn.cluster for the K-Means algorithm.

python

import numpy as np
from sklearn.cluster import KMeans
import matplotlib.pyplot as plt

• Step 2: Prepare the Data

For this example, we’ll use a simple dataset of points to be clustered. Each data point has two features.

python

# Sample data (X, Y coordinates)
data = np.array([
    [1, 2], [2, 3], [3, 4], [8, 8], [8, 9], [25, 80], [24, 79], [25, 77]
])

• Step 3: Set Up and Fit the K-Means Model

We’ll create an instance of KMeans with n_clusters=3, meaning we want to split the data into 3 clusters. Then, we fit the model to our data.

python

# Initialize K-Means with 3 clusters
kmeans = KMeans(n_clusters=3, random_state=0)

# Fit the model to the data
kmeans.fit(data)

• Step 4: Get the Cluster Labels and Centroids

After fitting the model, we can retrieve the cluster labels for each data point and the centroids of the clusters.

python

# Get cluster labels for each data point
labels = kmeans.labels_

# Get coordinates of cluster centroids
centroids = kmeans.cluster_centers_

print("Cluster Labels:", labels)
print("Cluster Centroids:", centroids)

Output:

lua

Cluster Labels: [0 0 0 1 1 2 2 2]
Cluster Centroids: [[ 2.         3.        ]
                    [ 8.         8.5       ]
                    [24.66666667 78.66666667]]

• Cluster Labels: Each data point is assigned a label indicating which cluster it belongs to (0, 1, or 2).
• Cluster Centroids: The coordinates of the centers of each cluster.
   

• Step 5: Visualize the Clusters

To better understand the results, we can plot the data points, their assigned clusters, and the centroids.

python

# Plot data points with colors based on cluster label
plt.scatter(data[:, 0], data[:, 1], c=labels, cmap='viridis', label='Data Points')
plt.scatter(centroids[:, 0], centroids[:, 1], s=200, c='red', marker='X', label='Centroids')

plt.title("K-Means Clustering")
plt.xlabel("Feature 1")
plt.ylabel("Feature 2")
plt.legend()
plt.show()

Explanation of Results

• Data Points: Each point is colored according to its cluster label, showing which points belong together.
• Centroids: The red "X" markers represent each cluster's centroids, or centers. Points are grouped around their respective centroids, showing the natural separation of the data.

Advantages of Cluster Analysis in Data Mining

• Clustering reveals hidden patterns and groups in data, making it easier to understand.
• Clusters can help predict trends and behaviors based on historical groupings.
• Businesses can use clustering to target customer groups effectively, improving marketing and personalization.
• Clustering supports decision-making in areas like risk assessment and targeted advertising by clarifying key groups.
• Clustering reduces data complexity by organizing large datasets into manageable groups for analysis.

Disadvantages of Cluster Analysis in Data Mining

• Clustering results can vary, leading to subjective interpretations depending on the method used.
• Some methods, like K-Means, require pre-set parameters (e.g., number of clusters) that can affect outcomes.
• Clustering may struggle with outliers or noisy data, which can distort results.

How Companies Use Cluster Analysis to Make Better Predictions

Cluster analysis helps companies find patterns and make informed predictions. Here’s how it’s used in different industries:

• Retail Industry: Improving Customer Segmentation

  • Intro: Retailers want to personalize marketing to boost sales and customer loyalty.
  • Challenge: With large customer bases, sending generic offers reduces engagement.
  • Solution: Clustering groups customers by buying habits and preferences. 75% of retail companies use clustering to improve customer segmentation, which enables personalized offers that drive higher engagement.

• Healthcare: Grouping Patients by Medical Histories

  • Intro: Healthcare providers aim to identify at-risk patients to improve care.
  • Challenge: High readmission rates strain hospital resources.
  • Solution: Clustering groups patients with similar health profiles, which makes it easier to spot those at risk. This approach has helped hospitals lower readmissions by focusing on targeted care.

• Financial Services: Identifying High-Risk Customers

  • Intro: Banks need accurate credit assessments to avoid defaults.
  • Challenge: Traditional credit scores don’t always capture customer risk accurately.
  • Solution: Clustering identifies high- and low-risk customers by income, spending, and credit history. Banks have reduced loan defaults by 15% by using clustering to refine risk models.

Master Cluster Analysis with upGrad’s Data Science Program

Learn cluster analysis, real-world applications, and essential tools for a data-driven career.

What You’ll Learn

• Clustering Basics: Techniques like Partitioning, Hierarchical, and Density-Based Clustering.
• Hands-On Projects: Work with Python and sklearn on real datasets.

Real-World Applications

• Customer Segmentation: Group customers for targeted marketing.
• Healthcare Clusters: Identify patient groups for tailored care.
• Financial Analysis: Cluster by risk for smarter credit scoring.

Skills You’ll Develop

• Data Pre-Processing
• Choosing Clustering Techniques
• Algorithm Proficiency: Practice with k-means, DBSCAN, and more.

Tools & Technologies

• Programming: Python & SQL
• Visualization: Tableau, Power BI
• Libraries: pandas, matplotlib

Why upGrad?

• Global Partnerships: Learn with IIIT-B and more.
• Industry Projects: Real-world case studies for job-ready skills.
• Flexible Learning: Online with 24/7 mentor support.

Start learning with upGrad today!

Elevate your expertise with our range of Popular Data Science Courses. Browse the programs below to discover your ideal fit.

Advance your in-demand data science skills with our top programs. Discover the right course for you below.

Explore popular articles related to data science to enhance your knowledge. Browse the programs below to find your ideal match.