[Day 19] Unsupervised Machine Learning Type 2 - Hierarchical Clustering (with a Small Python Project)

🎯 What is Hierarchical Clustering?

Hierarchical Clustering is a clustering algorithm that groups similar data points based on hierarchy rather than a fixed number of clusters like K-Means. It builds a tree-like structure (dendrogram) to show how clusters are formed at different levels.

Example: Borrower Segmentation in Credit Risk Analysis

Imagine a bank that wants to segment borrowers based on their creditworthiness. The bank collects data on credit score, income, loan amount, interest rate, and repayment history.

🔹 Hierarchical Clustering helps!

• It groups borrowers into low-risk, medium-risk, and high-risk based on financial behavior.
• Borrowers in Cluster 0 have the highest credit scores and stable income.
• Borrowers in Cluster 1 have lower credit scores and higher loan defaults.
• Helps banks in loan approvals and risk assessment.

🛒 Real-World Use Cases of Hierarchical Clustering

1.Customer Segmentation (E-Commerce & Retail)

📌 Problem: A business wants to group customers based on their shopping preferences. 🎯 Hierarchical Clustering Solution:

• Creates a hierarchy of customer groups (e.g., Budget Shoppers → Mid-range Shoppers → Premium Shoppers).
• Helps businesses refine their marketing strategies.

2.Fraud Detection in Banking

📌 Problem: A bank needs to detect fraudulent transactions. 🎯 Hierarchical Clustering Solution:

• Groups transactions into low-risk, medium-risk, and high-risk based on spending patterns.
• Helps banks like JP Morgan and Citibank identify emerging fraud patterns.

3.Medical Research: Classifying Cancer Cells

📌 Problem: Doctors need to classify cancerous and non-cancerous cells. 🎯 Hierarchical Clustering Solution:

• Groups cells into healthy, pre-cancerous, and cancerous based on genetic similarities.
• Used in hospitals for early cancer detection.

📊 How Hierarchical Clustering Works (Step by Step)

Step 1: Start with Individual Clusters

• Each data point is initially treated as its own cluster.

Step 2: Merge the Closest Clusters

• The two most similar clusters are merged into one.

Step 3: Repeat Until One Big Cluster is Left

• This merging process continues until all points belong to a single cluster.

Step 4: Use the Dendrogram to Decide Clusters

• A dendrogram (tree diagram) is used to cut at the optimal level to form meaningful clusters.

🎯 Final Result: A hierarchy of clusters, allowing flexibility in choosing the number of groups.

🖥 Python Mini Project: Borrower Segmentation for Credit Risk

Let’s build a Hierarchical Clustering model to segment borrowers based on their credit profiles.

🔹 Dataset: borrower_data.csv

Save this table as borrower_data.csv

📝 Python Code for Hierarchical Clustering (Borrower Segmentation)

import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import scipy.cluster.hierarchy as sch
from sklearn.cluster import AgglomerativeClustering

# Step 1: Load Dataset
data = pd.read_csv("borrower_data.csv")
X = data[['Credit Score', 'Income ($)', 'Loan Amount ($)', 'Interest Rate (%)', 'Repayment History']]

# Step 2: Create Dendrogram to Find Optimal Clusters
plt.figure(figsize=(12, 6))
dendrogram = sch.dendrogram(sch.linkage(X, method='ward'))
plt.xlabel('Borrowers')
plt.ylabel('Distance')
plt.title('Dendrogram for Borrower Segmentation')
plt.grid()
plt.show()

# Step 3: Apply Hierarchical Clustering
hc = AgglomerativeClustering(n_clusters=3, metric='euclidean', linkage='ward')
data['Cluster'] = hc.fit_predict(X)

# Step 4: Visualize Clusters in a Pairplot
sns.pairplot(data, hue='Cluster', palette='Set1')
plt.show()

# Step 5: Scatter Plot with Multiple Features
plt.figure(figsize=(10, 6))
sns.scatterplot(x=data['Credit Score'], y=data['Income ($)'], hue=data['Cluster'], palette='Set1', s=100)
plt.xlabel('Credit Score')
plt.ylabel('Income ($)')
plt.title('Borrower Segmentation using Hierarchical Clustering')
plt.legend(title='Cluster')
plt.grid()
plt.show()

Result:

📊 Explanation of the Borrower Segmentation Chart

This scatter plot represents borrower segmentation using Hierarchical Clustering, where:

• X-axis (Credit Score): Represents the borrower's credit score (higher scores indicate better creditworthiness).
• Y-axis (Income $): Represents the annual income of the borrower.
• Color-coded Clusters: Each borrower is assigned to a cluster based on their financial characteristics.

🔹 Interpreting the Clusters

1. Cluster 0 (Red Points - Medium-Risk Borrowers)
   
   • Most borrowers in this cluster have moderate credit scores (650-750).
   • They also have decent income levels but may not be the highest earners.
   • These borrowers are likely to get loans but with slightly higher interest rates.
     
2. Cluster 1 (Blue Points - High-Risk Borrowers)
   
   • These borrowers have low credit scores (below 650) and lower income levels.
   • They may have poor repayment history or high debt burdens.
   • They are considered high-risk for banks, meaning they might either get denied loans or charged higher interest rates.
     
3. Cluster 2 (Green Points - Low-Risk Borrowers)
   • Borrowers in this cluster have very high credit scores (above 780).
   • They also have higher income levels (above $100,000).
   • These individuals are low-risk, meaning banks can offer them better loan terms with lower interest rates.
     

🎯 What Can We Conclude?

• Banks can use this segmentation to make loan approval decisions.
• Cluster 2 borrowers are the best candidates for loans.
• Cluster 1 borrowers need financial improvement before getting favorable loans.
• Cluster 0 borrowers fall somewhere in between and may receive standard loan terms.

This approach helps financial institutions optimize risk assessment while offering loans to different borrower profiles! 🚀

💡 Nutshell

✅ Hierarchical Clustering is useful when we don’t know the number of clusters in advance.
✅ It helps banks in credit risk assessment and loan approvals.
✅ It is widely used in banking, healthcare, and customer segmentation.

💬 Join the DecodeAI WhatsApp Channel for regular AI updates → Click here