🛍️ Customer Segmentation using Unsupervised Machine Learning

Segmenting mall customers into behavior-based groups using K-Means clustering, PCA, and standardization — turning raw demographic and spending data into actionable marketing insights.

---

📌 Project Overview

Retailers often assume that high income = high value customer — but is that actually true? This project uses unsupervised machine learning to answer that question empirically, by clustering mall customers based on their age, income, and spending behavior, without any pre-labeled categories.

The goal: give a business a data-driven way to understand who their customers actually are, instead of relying on assumptions.

🎯 Objective

Apply multi-variable clustering to group customers by behavior, using:

• Standardization to normalize feature scales
• Principal Component Analysis (PCA) to reduce dimensionality for visualization
• Elbow Method and Silhouette Score to determine the optimal number of clusters
• K-Means Clustering to segment customers into distinct behavioral groups

📊 Dataset

Source: Mall Customer Segmentation Dataset (Kaggle)

Column	Description
CustomerID	Unique identifier (excluded from clustering)
Gender	Customer gender
Age	Customer age
Annual Income (k$)	Annual income in thousands of dollars
Spending Score (1-100)	Score assigned by the mall based on spending behavior

200 records, no missing values, no significant outliers.

🔍 Exploratory Data Analysis

Key finding before any modeling: Annual Income and Spending Score are essentially uncorrelated (r ≈ 0.01).

This is the central insight that justifies the entire approach — if income predicted spending, a simple business rule ("target high earners") would be enough. Since it doesn't, customers must be segmented using a method that can find non-obvious behavioral patterns — which is exactly what clustering does.

⚙️ Methodology

1. Standardization

Features were scaled using StandardScaler (mean = 0, std = 1). This is essential for K-Means because it's a distance-based algorithm — without scaling, Annual Income (range: 15–137) would dominate Age (range: 18–70) purely due to its larger numeric range, regardless of actual importance.

2. Dimensionality Reduction (PCA)

Reduced 3 standardized features into 2 principal components for visualization, retaining 77.6% of total variance — enough to meaningfully visualize cluster separation in 2D.

3. Optimal K Selection

Tested K = 2 to 10 using both the Elbow Method (inertia/WCSS) and Silhouette Score:

K = 6 was selected — it produced the highest silhouette score (0.428), and the elbow curve visibly flattens beyond this point, meaning additional clusters stop providing meaningful separation.

4. K-Means Clustering

Final model fit with K = 6, random_state=42 for reproducibility.

📈 Results — Customer Segments

Cluster	Avg. Age	Avg. Income	Avg. Spending Score	Size	Segment Label
0	56	$54k	49	45	Older, Moderate Spenders
1	27	$57k	48	39	Young, Average Spenders
2	42	$89k	17	33	High Income, Low Spenders
3	33	$87k	82	39	High Income, High Spenders (VIP)
4	25	$25k	78	23	Low Income, High Spenders
5	46	$26k	19	21	Low Income, Low Spenders

💡 Business Recommendations

• Cluster 3 (VIP) — Highest-value segment. Prioritize loyalty programs, early access to sales, and premium experiences to retain them.
• Cluster 2 (High Income, Low Spend) — The biggest missed opportunity. These customers can afford to spend more but aren't engaging — worth investigating with targeted, premium-brand offers.
• Cluster 4 (Low Income, High Spend) — Enthusiastic but budget-constrained. Retain with discounts, installment options, and loyalty points; monitor for churn risk if competitors offer cheaper alternatives.
• Cluster 5 (Low Income, Low Spend) — Lowest marketing priority; acquisition cost is unlikely to be justified by return.
• Clusters 0 & 1 — Stable, average-value segments suited to general seasonal promotions rather than targeted campaigns.

🛠️ Tech Stack

• Python 3
• pandas / numpy — data manipulation
• matplotlib / seaborn — visualization
• scikit-learn — StandardScaler, PCA, KMeans, silhouette_score

📁 Project Structure

customer-segmentation/
├── data/
│   └── raw/
│       └── Mall_Customers.csv
├── notebooks/
│   └── 01_eda.ipynb
├── src/
│   └── data_loader.py
├── outputs/
│   ├── customers_with_clusters.csv
│   └── figures/
├── requirements.txt
└── README.md

▶️ How to Run

git clone https://github.com/DataWithHamza/Customer-Segmentation.git
cd Customer-Segmentation
pip install -r requirements.txt

Then open notebooks/01_eda.ipynb in VS Code or Jupyter and run all cells.

🚀 Future Improvements

• Engineer RFM (Recency, Frequency, Monetary) features from raw transaction-level data for a more realistic segmentation
• Compare K-Means against Hierarchical Clustering and DBSCAN
• Build an interactive dashboard (Streamlit/Power BI) for stakeholders to explore segments
• Deploy as an API that assigns new customers to a segment in real time

👤 Author

Hamza — Data Science Student & ML Intern 📧 hamza.professional.connect@gmail.com 🔗 GitHub

---

This project was built as a hands-on portfolio piece to demonstrate the full unsupervised ML workflow: from raw data to business-actionable customer segments.