Back to All Projects
Machine Learning & Customer AnalyticsCase Study • Technical Overview
Customer Behavioral Segmentation & Churn Prevention
A machine learning pipeline that categorizes e-commerce customers based on Recency, Frequency, and Monetary (RFM) metrics, paired with unsupervised clustering algorithms to tailor retention campaigns.
PythonScikit-LearnK-MeansPandasPlotlySeabornSQL
0.89
AUC-ROC
-6.8%
Churn Cut
4.2x
CTR Lift
$185K
Revenue Saved
1. Business Problem & Objective
Generic marketing emails led to low engagement (1.8% open rate) and elevated annual customer churn (22%), straining customer acquisition cost budgets.
2. Dataset Description
850,000 transaction records across 120,000 unique registered customer accounts over a 3-year period.
3. Data Cleaning & Preprocessing
- Calculated RFM metrics per individual customer account based on baseline snapshot date.
- Applied Log transformation and StandardScaler normalization to correct right-skewed monetary distributions.
- Removed zero-value transactions and guest checkout anomalies.
4. Methodology & Machine Learning Architecture
Analytical Steps
- Elbow Method & Silhouette Analysis to determine optimal cluster count (K=5).
- K-Means clustering & Principal Component Analysis (PCA) for 2D/3D cluster visualization.
- Logistic Regression & Random Forest modeling to classify churn risk probability.
Model Development & Evaluation
Trained Random Forest Classifier achieving an AUC-ROC of 0.89 for predicting customer churn risk 30 days prior to account inactivity.
5. Results & Strategic Business Insights
Key Results Achieved
- Identified 'At-Risk Champions' cohort generating 35% of revenue.
- Targeted win-back email campaign increased email CTR by 4.2x.
- Reduced annual customer churn rate by 6.8 percentage points.
Actionable Insights
- Customers who purchase twice within the first 14 days have an 82% higher 2-year LTV.
- Offering proactive support outreach to 'High-Monetary / Low-Recency' users rescued $185K in impending churn revenue.