Clustering Analysis on Wholesale Customers Dataset

This project performs a comparative performance study of clustering algorithms using the Wholesale Customers dataset from the UCI Machine Learning Repository. The analysis leverages different preprocessing techniques, varying cluster sizes, and multiple evaluation metrics to determine the most effective clustering configuration.

---

📊 Dataset Overview

• Name: Wholesale Customers Dataset
• Source: UCI Machine Learning Repository
• Number of Features: 7
• Number of Records: 440
• Description: The dataset contains annual spending in monetary units on various product categories for customers from a wholesale distributor.

---

⚙️ Methodology

💡 Clustering Algorithms Used

• K-Means
• Hierarchical Clustering (HCLUST)
• MeanShift

🔁 Preprocessing Techniques

• No Processing
• Normalization
• Transformation
• PCA
• Transformation + Normalization (T+N)
• Transformation + Normalization + PCA (T+N+PCA)

🔢 Cluster Counts

• 3 clusters
• 4 clusters
• 5 clusters

📐 Evaluation Metrics

• Silhouette Score
• Calinski-Harabasz Index
• Davies-Bouldin Score

---

✅ Results Summary

Metric	Best Value
Best Algorithm	MEANSHIFT
Best Clusters	3
Best Silhouette	0.9076

All evaluations were performed using the PyCaret library.

---

📊 Visualizations

All model evaluations are also visualized using grouped bar plots for:

• Silhouette Score
• Calinski-Harabasz Index
• Davies-Bouldin Score

Each model's performance across different preprocessing techniques and cluster sizes is presented.

---

📁 Files Included

• clustering_results.csv – Final result table with all configurations
• Saved plots for each metric/model
• Jupyter Notebook / Colab Notebook for reproducibility