Customer Segmentation using DBSCAN & PCA (Unsupervised Learning Project)

Performed an advanced unsupervised learning analysis to identify meaningful customer segments based on booking behavior.

The preprocessing pipeline included:

• Missing value handling

• Duplicate removal

• Outlier detection using IQR

• One-Hot Encoding

• Feature scaling using StandardScaler

• Dimensionality reduction using PCA

Two clustering algorithms were evaluated:

• K-Means

• DBSCAN

Evaluation Metrics:

• Silhouette Score

• Davies–Bouldin Index

Results:

DBSCAN significantly outperformed K-Means by producing more compact and well-separated clusters (Silhouette ≈ 0.247 vs 0.048). It also successfully detected noise points and handled irregular cluster shapes.

This project demonstrates practical expertise in clustering evaluation, dimensionality reduction, and real-world customer behavior analysis.