ML-Clustering

🧠 Unsupervised Learning — Customer Segmentation with Clustering

Date: April 24, 2024
Author: Clement Airohuodion

---

🔍 View Full Project Files

This project demonstrates how clustering — a powerful unsupervised learning technique — can be used to identify customer segments in the competitive world of online retail, specifically in the footwear industry. The goal is to support data-driven decisions in marketing, credit strategy, and customer retention.

Using a combination of K-Means and Hierarchical Clustering, this analysis reveals meaningful insights into customer behaviors, spending patterns, and demographic trends.

---

📊 Dataset

• Source: Kaggle - Online Shop Dataset
• Size: 8,950 entries × 18 features
• Fields: Balance, Purchases, Credit Limit, Payments, Tenure, and more
• Privacy: Fully anonymized and GDPR-compliant

---

Figure 1: This code reads and inspects the dataset and showing first 5 rows using “.head()”.

---

🧹 Preprocessing

1. Missing Values: Handled using mean imputation
2. Outliers: Removed using IQR method
3. Standardization: Applied StandardScaler for uniform scale
4. Data Visualizations:
   • Pairplots & histograms to explore feature distribution
   • Interactive scatter plots & heatmaps using plotly
   • Boxplots & KDEs to understand payment and balance behaviors

---

Figure 2: The code above creates a pairplot to visualize some relationships amongst key features like Balance, Purchases, Credit_limit, Payments, Minimum_payments.

---

Figure 3: Above is the correlation matrix for raw data, where the correlation between Balance and Cash deposit is “0.33”.

---

Figure 4: Above is the correlation matrix for cleaned data, where the correlation between Balance and Cash deposit is “0.50”.

---

🧠 Algorithms Applied

✅ K-Means Clustering

• Elbow method used to identify optimal clusters (K = 3)
• PCA reduced data to 2D for visualization
• Key Clusters:
  • High-spending customers
  • Discount seekers
  • Cash-advance-dependent segments

---

Figure 5: Above shows visualisation of the uniquely coloured clusters generated by K-Means, also the centroids are highlighted.

---

✅ Hierarchical Clustering

• Ward’s method used
• Dendrogram helped visualize group formation
• Showed evolving behavioral clusters and niche segments

---

Figure 6: The Above dendogram shows the result of hierarchical merging using Ward’s method.

---

Figure 7: Visualisation of Hierarchical clustering in 2D scatter pot.

---

📈 Results & Insights

• Silhouette Score: 0.37 (Moderate clustering quality)
• Customer Segments:
  • Identified 3 major clusters with distinct spending and payment patterns
  • Cross-cluster comparison revealed actionable strategies for:
    • Loyalty rewards
    • Risk mitigation (credit stress)
    • Personalized promotions

---

📌 Conclusion

Clustering enabled strategic segmentation of customers, unlocking critical insights to:

• Enhance customer experience
• Improve credit risk management
• Enable targeted marketing

By integrating both K-Means and Hierarchical Clustering, this project provides a comprehensive view of customer behavior in online retail, aligning with real business challenges.

---

📁 Click to See Supporting Code & Notebooks

📧 Contact: C.O.Airohuodion@edu.salford.ac.uk
🔗 LinkedIn: linkedin.com/in/yourprofile
🔗 GitHub: github.com/Clemobrain

---