Customer Segmentation — K-Means & Hierarchical Clustering

Unsupervised learning project to identify distinct customer groups from the Mall Customers dataset using two clustering algorithms, with full evaluation and business interpretation.

---

Problem Statement

Retail businesses need to understand their customers' spending behaviour to design targeted marketing strategies. This project segments customers based on annual income and spending score to uncover natural groupings without predefined labels.

---

Dataset

Property	Detail
Source	Mall Customers Dataset — Kaggle
Records	200
Features used	Annual Income (k$), Spending Score (1–100)

---

Approach

• Exploratory Data Analysis — distributions, boxplots, scatter plots, correlation heatmap
• Feature selection — Age dropped due to weak correlation with Income and SpendScore
• Scaling — StandardScaler for K-Means, MinMaxScaler for Hierarchical
• Optimal k selection — Elbow method + KneeLocator + Silhouette score
• Clustering — K-Means and Agglomerative (Ward linkage) both with k=5
• Evaluation — ARI, Davies-Bouldin score, Calinski-Harabasz score
• Visualisation — Scatter plots, treemaps, pie charts, radar profiles, 3D interactive plot

---

Results

Metric	K-Means	Hierarchical
Davies-Bouldin Score	lower = better	lower = better
Calinski-Harabasz Score	higher = better	higher = better
Adjusted Rand Index	—	compared against K-Means

Both algorithms independently identified 5 customer segments with consistent groupings.

Cluster Personas (K-Means)

Cluster	Persona	Avg Income	Avg Spend Score
0	Balanced Middle	$55k	49
1	Premium Loyalists	$87k	82
2	Impulsive Spenders	$26k	79
3	Conservative Wealthy	$88k	17
4	Careful Savers	$26k	21

---

Tech Stack

Python · pandas · numpy · scikit-learn · scipy · matplotlib · seaborn · squarify · kneed · plotly · Flask · joblib

---

Project Structure

customer-segmentation/
├── income-spend-cluster-analysis.ipynb
├── Mall_Customers.csv
├── app.py
├── templates/
│   └── index.html
├── artifacts/
│   ├── model.pkl
│   └── scaler.pkl
├── requirements.txt
└── README.md

---

Setup

pip install -r requirements.txt
jupyter notebook income-spend-cluster-analysis.ipynb

---

Deployment

The project includes a Flask web application that serves the trained K-Means model as an interactive prediction API.

How it works

1. The trained K-Means model and StandardScaler are serialised via joblib and saved to the artifacts/ directory.
2. app.py loads these artifacts at startup and exposes two routes:
   • GET / — renders the input form
   • POST /predict — accepts Annual Income and Spending Score, scales the input, runs inference, and returns the predicted customer segment.

Run the web app locally

pip install flask joblib numpy
python app.py

Then open http://localhost:5000 in your browser.

Customer Segments served by the app

Cluster	Customer Type	Segment Name
0	Regular Customer	Average Customers
1	Premium Customer	High Income — High Spending
2	Value-Oriented Customer	Low Income — High Spending
3	Conservative Customer	High Income — Low Spending
4	Budget Customer	Low Income — Low Spending