Project 5: Customer Segmentation

Segment customers into groups based on behavior and demographics.

Starter code: Run starter.py after placing the segmentation CSV in data/.

Difficulty

Intermediate

Time Estimate

3-4 days

Skills You'll Practice

• Unsupervised Learning
• Clustering
• Dimensionality Reduction
• Data Visualization

Learning Objectives

By completing this project, you will learn to:

• Apply clustering algorithms
• Choose optimal number of clusters
• Interpret cluster results
• Use dimensionality reduction for visualization
• Extract business insights from clusters

Dataset

Mall Customer Segmentation

• Kaggle Mall Customer Data
• Customer demographics and spending
• Features: Age, Income, Spending Score

Project Steps

Step 1: Load and Explore Data

• Load dataset
• Explore feature distributions
• Check for missing values
• Visualize relationships

Step 2: Data Preprocessing

• Handle missing values
• Scale features (important for clustering!)
• Feature selection
• Check for outliers

Step 3: Dimensionality Reduction

• Apply PCA for visualization
• Use t-SNE for visualization
• Reduce dimensions if needed
• Visualize in 2D/3D

Step 4: K-Means Clustering

• Determine optimal k (elbow method)
• Apply K-Means
• Analyze clusters
• Visualize clusters

Step 5: Hierarchical Clustering

• Apply hierarchical clustering
• Create dendrogram
• Choose number of clusters
• Compare with K-Means

Step 6: DBSCAN

• Apply DBSCAN
• Handle outliers
• Compare with other methods

Step 7: Cluster Analysis

• Profile each cluster
• Identify cluster characteristics
• Business interpretation
• Actionable insights

Expected Deliverables

1. Jupyter Notebook with clustering analysis
2. Cluster Profiles describing each segment
3. Visualizations of clusters
4. Business Recommendations based on segments

Evaluation Metrics

• Silhouette Score: Cluster quality
• Inertia: Within-cluster sum of squares
• Business Metrics: Actionability of segments

Clustering Methods to Try

1. K-Means: Most common, fast
2. Hierarchical: Tree-like structure
3. DBSCAN: Density-based, handles outliers
4. Gaussian Mixture: Probabilistic

Tips

• Always scale features before clustering
• Use elbow method to find optimal k
• Visualize clusters in 2D/3D
• Interpret clusters for business
• Try multiple algorithms and compare
• Use silhouette score for evaluation

Resources

• Kaggle Customer Segmentation
• Scikit-learn Clustering
• Clustering Guide

Extensions

• Multiple clustering algorithms comparison
• Real-time customer segmentation
• Segment-based marketing strategies
• Dashboard for segment analysis

Next Steps

After completing this project:

• Try larger customer datasets
• Experiment with advanced clustering
• Move to Project 6: Time Series Forecasting