ML-powered customer analytics

Project Overview

This project demonstrates the development of an unsupervised ML algorithm that segments customers based on lifetime value (LTV) and provides actionable business insights. The goal is to help e-commerce companies identify high-value customers, understand purchasing behavior, and inform personalized marketing strategies.

The platform uses the Online Retail Dataset (UCI Machine Learning Repository) and was developed and deployed as a proof-of-concept in one day.

Dataset

• Source: UCI Online Retail Dataset
• Features:

Column	Description
InvoiceNo	Unique invoice number
StockCode	Unique product code
Description	Product description
Quantity	Number of items purchased
InvoiceDate	Date and time of invoice
UnitPrice	Price per unit
CustomerID	Unique customer identifier
Country	Customer country

• Data preprocessing steps:

  • Removed rows without CustomerID
  • Removed negative UnitPrice rows
  • Aggregated at customer-level to compute RFM metrics: Recency, Frequency, Monetary (TotalRevenue)

---

Methodology

1. Customer Segmentation

   • Calculated RFM metrics:

     • Recency: Days since last purchase
     • Frequency: Number of invoices/orders
     • Monetary: Total spend

   • Due to heavy skew in the Monetary and Frequency distributions, standard IQR-based outlier detection was ineffective. Instead, thresholds were set empirically, guided by Elbow and Silhouette analysis, to ensure well-separated, meaningful customer segments.

   • Applied log transformation and standard scaling

   • Used KMeans clustering to identify 3 segments:

     • VIP (Blue)
     • Middle Class (Turquoise)
     • Inactive (Brown)

2. KPIs Computed per Segment

   • Total Revenue per Segment
   • Average Order Value (AOV)
   • Purchase Frequency
   • Recency
   • Segment Size
   • Revenue Contribution (%)

See the Analytics Dashboard on Tableau

---

Key Findings

Segment	% of Customers	% of Revenue	Notes
VIP	21%	57%	High frequency, high spend. Most profitable segment.
Middle Class	40%	35%	Highest AOV, lower frequency than VIP. Opportunity for upselling.
Inactive	39%	8%	Rarely purchase; low revenue contribution.

Insights:

• 61% of customers (VIP + Middle Class) generate 92% of total revenue
• Middle Class customers have higher AOV but half the frequency of VIPs
• Targeted campaigns should focus on retention and upsell for VIP and Middle Class customers
• Inactive segment can be targeted with win-back campaigns or cost-efficient marketing strategies

---

How to Run

1. Clone the repository:

git clone https://github.com/JavieraAlmendrasVilla/Customer-Segmentation.git

2. Install dependencies:

pip install -r requirements.txt

3. Run the notebook:

jupyter notebook customer_segmentation.ipynb

---

Technologies Used

• Python (pandas, numpy, matplotlib, seaborn, scikit-learn)
• Jupyter Notebook
• KMeans clustering for segmentation
• PCA for visualization

---

Portfolio Takeaway

This project demonstrates the ability to:

• Preprocess and clean real-world e-commerce data
• Compute RFM metrics and detect outliers
• Segment customers with unsupervised learning
• Derive actionable business insights
• Visualize results clearly for stakeholders

---