Customer Segmentation with K-Means Clustering

01Loading the dataset

We'll use the Online Retail dataset from the UCI Machine Learning Repository — real transactional data from a UK-based online retailer, covering invoices, products, quantities and customer IDs over one year.

import pandas as pd
import numpy as np

# UCI Online Retail dataset (real transactional data)
url = "https://archive.ics.uci.edu/ml/machine-learning-databases/00352/Online%20Retail.xlsx"
df = pd.read_excel(url)

print(df.shape)
print(df.head())
print(df['CustomerID'].nunique(), "unique customers")

02Exploratory data analysis

For customer segmentation, we don't analyse raw transactions — we engineer RFM features (Recency, Frequency, Monetary value), the industry-standard way to describe customer behaviour for clustering.

import matplotlib.pyplot as plt
import seaborn as sns
from datetime import timedelta

# Clean: remove cancelled orders and missing customer IDs
df = df.dropna(subset=['CustomerID'])
df = df[df['Quantity'] > 0]
df['TotalPrice'] = df['Quantity'] * df['UnitPrice']

# Engineer RFM features per customer
snapshot_date = df['InvoiceDate'].max() + timedelta(days=1)

rfm = df.groupby('CustomerID').agg({
    'InvoiceDate': lambda x: (snapshot_date - x.max()).days,  # Recency
    'InvoiceNo': 'nunique',                                    # Frequency
    'TotalPrice': 'sum'                                        # Monetary
})
rfm.columns = ['Recency', 'Frequency', 'Monetary']

print(rfm.describe())

# Distribution of each RFM feature
fig, axes = plt.subplots(1, 3, figsize=(15, 4))
for ax, col in zip(axes, rfm.columns):
    sns.histplot(rfm[col], ax=ax, color='#1a4a3a')
    ax.set_title(f'{col} Distribution')
plt.savefig('rfm_distributions.png')

03Preprocessing & the elbow method

K-Means needs us to choose K (the number of clusters) in advance. The elbow method helps us pick a sensible K by plotting inertia (within-cluster variance) against different values of K and looking for the point where adding more clusters stops helping much.

from sklearn.preprocessing import StandardScaler
from sklearn.cluster import KMeans

# Scale RFM features — essential since they're on very different ranges
scaler = StandardScaler()
rfm_scaled = scaler.fit_transform(rfm)

# Try K from 2 to 10, record inertia for each
inertias = []
K_range = range(2, 11)
for k in K_range:
    km = KMeans(n_clusters=k, random_state=42, n_init=10)
    km.fit(rfm_scaled)
    inertias.append(km.inertia_)

plt.figure(figsize=(8, 5))
plt.plot(K_range, inertias, marker='o', color='#1a4a3a')
plt.xlabel('Number of Clusters (K)')
plt.ylabel('Inertia')
plt.title('Elbow Method for Optimal K')
plt.savefig('elbow_method.png')

04Training the model

With K=4 chosen, we fit the final K-Means model and assign each customer to a cluster.

# Train final model with K=4
kmeans = KMeans(n_clusters=4, random_state=42, n_init=10)
rfm['Cluster'] = kmeans.fit_predict(rfm_scaled)

# Examine each cluster's average RFM profile
cluster_profile = rfm.groupby('Cluster').agg({
    'Recency': 'mean',
    'Frequency': 'mean',
    'Monetary': 'mean',
    'Cluster': 'count'
}).rename(columns={'Cluster': 'CustomerCount'})

print(cluster_profile)

Cluster	Recency (days)	Frequency	Monetary (£)	Segment name
0	18	12.4	5,820	Champions — recent, frequent, big spenders
1	245	1.8	410	Lost customers — haven't purchased in months
2	42	5.1	1,650	Loyal customers — steady, moderate spenders
3	89	2.3	720	At-risk — used to buy, slowing down

05Evaluation & visualisation

Since there's no ground truth label, we evaluate clustering differently — using the Silhouette Score (how well-separated clusters are) and visualising the clusters with PCA, which compresses our 3 RFM dimensions down to 2D for plotting.

from sklearn.metrics import silhouette_score
from sklearn.decomposition import PCA

silhouette = silhouette_score(rfm_scaled, rfm['Cluster'])
print(f"Silhouette Score: {silhouette:.4f}")

# Reduce to 2D for visualization
pca = PCA(n_components=2)
rfm_pca = pca.fit_transform(rfm_scaled)

plt.figure(figsize=(9, 7))
scatter = plt.scatter(rfm_pca[:, 0], rfm_pca[:, 1],
                       c=rfm['Cluster'], cmap='Greens', alpha=0.6)
plt.xlabel('PCA Component 1')
plt.ylabel('PCA Component 2')
plt.title('Customer Segments (PCA-Reduced)')
plt.colorbar(scatter, label='Cluster')
plt.savefig('customer_segments_pca.png')

06Real-world use case

This project demonstrates the unsupervised learning workflow: when you don't have labels, you can still extract genuinely useful structure from data — the RFM + K-Means combination here is one of the most widely deployed clustering techniques in retail analytics today.