Case Study Overview
Building automated data pipelines and machine learning classifiers to predict customer cancellations (churn) for a subscription-based digital platform.
Problem Statement
Proactively identify active subscribers who are likely to cancel their membership in the upcoming month.
Analytical Focus
Feature engineering on activity logs, training classification models, and mapping model parameters for business reviews.
The Analytical Approach
Using Python, I consolidated subscriber usage logs, transaction records, and client support cases. I engineered indicators for engagement degradation, such as decreased login frequency and payment delays.
I evaluated multiple classification algorithms (Logistic Regression, Decision Trees, Random Forest). Random Forest was selected due to higher accuracy and recall metrics.
Results & Business Value
Key Metric Improvement
12% Churn Reduction Achieved
- Enabled marketing teams to automatically target high-risk groups with retention emails.
- Discovered that support response times above 4 hours increase churn probability by 25%.
- Automated customer profiling dashboards that update retention indexes daily.