Unveiling Retail Dynamics: Mining Predictive Insights and Customer Segmentation from Online Retail Data

---

Project Overview

This project presents a comprehensive customer segmentation and predictive modeling pipeline using the UCI Online Retail Dataset. Leveraging RFM analysis, K-Means clustering, and supervised machine learning models (Logistic Regression, Random Forest, XGBoost, Deep Learning), it explores how businesses can identify, classify, and forecast high-value customers based on behavioral data.

The methodology is grounded in both academic research and practical relevance—supporting targeted marketing, CRM optimization, and long-tail customer value maximization in e-commerce settings.

---

Objectives

• Understand key purchasing patterns and product associations.
• Segment customers using RFM scores and K-Means clustering.
• Predict high-value customers using ML models based on behavioral signals.
• Provide actionable insights for targeted marketing and retention strategies.

---

Project Structure

Section	Description
1. Data Cleaning	Handled missing values, duplicate entries, cancellations, and pricing outliers.
2. Feature Engineering	Created Revenue, Recency, Frequency, InvoiceTime, IsWeekend, and CountryCode.
3. RFM Segmentation	Assigned RFM scores (1–5) per customer and grouped into strategic segments (e.g., Champions, At-Risk).
4. Clustering	Applied K-Means to RFM data (k=4), validated using Elbow and Silhouette methods, visualized via PCA.
5. Predictive Modeling	Built and compared 4 models (LogReg, RF, XGB, DL) to predict top 30% high-value customers.
6. Evaluation & Business Insights	Measured using Accuracy, ROC AUC, F1-score, and confusion matrix; translated findings into CRM strategies.

---

Key Technologies

• Languages: Python
• Libraries: pandas, numpy, matplotlib, seaborn, plotly, scikit-learn, geopandas, keras, xgboost
• ML Models:
  • Logistic Regression
  • Random Forest
  • XGBoost
  • Deep Learning (Neural Network)
• Evaluation Metrics: Accuracy, ROC-AUC, F1-score, Confusion Matrix

---

Key Findings

• 🎯 Champions (0.3%) generate 65%+ of total revenue—top priority for loyalty programs.
• 🔄 Cluster labels (from K-Means) significantly improved model performance.
• 🔍 PCA revealed interpretable clusters, confirming the validity of segmentation.
• 🤖 Deep Learning outperformed all models (ROC AUC: 0.932, F1-score: 0.78).
• 💡 High-value customers can be predicted using only Recency, Frequency, and Cluster ID—no monetary input needed.

---

Model Performance Summary

Metric	Logistic Regression	Random Forest	XGBoost	Deep Learning
Accuracy	0.869	0.867	0.867	0.874
ROC AUC	0.923	0.917	0.920	0.932
F1-score (Class 1)	0.76	0.76	0.76	0.78

✅ Deep Learning captured non-linear customer behaviors;
✅ Logistic Regression remains suitable for interpretable business reporting.

---

Business Impact

• 📈 Real-time prediction of customer value for CRM platforms
• 🛍️ Segment-specific marketing and re-engagement strategies
• 💰 Improved ROI through customer retention and behavior-driven personalization
• 🧩 Deployable models with high interpretability and generalization

---

Installation

To install the required packages:

pip install pandas numpy seaborn matplotlib scikit-learn plotly geopandas xgboost keras