Skip to main content

​
Overview

Unsupervised learning is a machine learning paradigm where algorithms discover patterns and structures in data without labeled examples. Unlike supervised learning, where we provide input-output pairs, unsupervised learning works with unlabeled data to find hidden insights.
This module corresponds to Module A7 of the bootcamp, focusing on customer segmentation using clustering techniques.

​
What is Unsupervised Learning?

Unsupervised learning techniques analyze data to identify:
  • Natural groupings or clusters
  • Hidden patterns and relationships
  • Dimensionality reduction for visualization
  • Anomaly detection

​
Key Use Cases

Based on the Retail Insights S.A. customer segmentation project:

Customer Segmentation

Group customers by behavior patterns to design personalized marketing campaigns

Market Basket Analysis

Discover purchasing patterns and product associations

Anomaly Detection

Identify unusual patterns or outliers in customer behavior

Data Exploration

Visualize high-dimensional data in 2D/3D space

​
Dimensionality Reduction

Dimensionality reduction techniques transform high-dimensional data into lower dimensions while preserving important information.

​
PCA (Principal Component Analysis)

PCA is a linear dimensionality reduction technique that:
  • Finds orthogonal axes (principal components) that maximize variance
  • Orders components by explained variance
  • Enables visualization in 2D or 3D
From the customer segmentation project:
PCA was applied to:
  • Analyze how much variance different principal components explain
  • Obtain a 2-dimensional representation (PC1 and PC2) for customer visualization
  • Reduce computational complexity for clustering algorithms
from sklearn.decomposition import PCA
import matplotlib.pyplot as plt

# Apply PCA to reduce to 2 dimensions
pca = PCA(n_components=2)
X_pca = pca.fit_transform(X_scaled)

# Visualize explained variance
print(f"Explained variance ratio: {pca.explained_variance_ratio_}")
print(f"Total variance explained: {sum(pca.explained_variance_ratio_):.2%}")

# Plot customers in 2D space
plt.figure(figsize=(10, 6))
plt.scatter(X_pca[:, 0], X_pca[:, 1], alpha=0.5)
plt.xlabel('PC1')
plt.ylabel('PC2')
plt.title('Customer Distribution in PCA Space')
plt.show()

​
t-SNE (t-distributed Stochastic Neighbor Embedding)

t-SNE is a non-linear technique particularly good for visualization:
  • Preserves local neighborhood relationships
  • Creates meaningful 2D/3D visualizations of complex data
  • Better at revealing cluster structure than PCA
From the project:
t-SNE was applied to obtain a 2D representation that preserves local relationships between customers (neighborhoods). This technique is excellent for visualization but should not be used for feature extraction in downstream models.
from sklearn.manifold import TSNE

# Apply t-SNE for visualization
tsne = TSNE(n_components=2, random_state=42, perplexity=30)
X_tsne = tsne.fit_transform(X_scaled)

# Visualize with cluster colors
plt.figure(figsize=(10, 6))
plt.scatter(X_tsne[:, 0], X_tsne[:, 1], c=cluster_labels, cmap='viridis', alpha=0.6)
plt.colorbar(label='Cluster')
plt.title('Customer Clusters Visualized with t-SNE')
plt.show()

​
Data Preprocessing for Unsupervised Learning

​
Handling Missing Values

From the project preprocessing steps: 
from sklearn.impute import SimpleImputer

# Numeric variables: impute with median
num_imputer = SimpleImputer(strategy='median')
X_numeric = num_imputer.fit_transform(df_numeric)

# Categorical variables: impute with most frequent value
cat_imputer = SimpleImputer(strategy='most_frequent')
X_categorical = cat_imputer.fit_transform(df_categorical)

​
Feature Scaling

Always scale features for distance-based unsupervised algorithms. The customer segmentation project used StandardScaler to center variables at mean 0 and standard deviation 1.
from sklearn.preprocessing import StandardScaler, OneHotEncoder

# Scale numeric features
scaler = StandardScaler()
X_numeric_scaled = scaler.fit_transform(X_numeric)

# One-hot encode categorical features
encoder = OneHotEncoder(handle_unknown='ignore')
X_categorical_encoded = encoder.fit_transform(X_categorical)

​
Business Insights

​
Customer Segments Identified

From the Retail Insights S.A. project analysis:
  • Profile: Middle-aged, occupations like Artist or Healthcare, mainly “Low” income level
  • Strategy: Economic bundles, simple loyalty programs, promotional campaigns
  • Profile: Executive and Lawyer occupations, “High” income levels
  • Strategy: Premium products, exclusive plans, VIP programs
  • Profile: Younger customers with lower seniority
  • Strategy: Onboarding programs, activation campaigns, personalized recommendations
  • Profile: Customers with less frequent or mixed patterns
  • Strategy: Individual analysis to detect opportunities or risks

​
Next Steps

After understanding unsupervised learning fundamentals, explore:

Clustering Algorithms

Learn K-Means, DBSCAN, and hierarchical clustering

Deep Learning

Move on to neural networks and deep learning fundamentals

​
Additional Resources

Build docs developers (and LLMs) love

Get started for freeTalk to us