Perceptive Analytics
Dec 19, 2023
No image
Customer Churn Predictive Model

Customer Churn Predictive Model

4-6 months
United States, San Francisco
view project
Service categories
Service Lines
Big Data
Domain focus
Consumer Products & Services
Big Data
Marketing Analytics


Data Complexity: Dealing with 11 years of diverse data, including membership details, transactions, and user logs, posed a challenge in extracting meaningful insights. Feature Engineering: Creating relevant features to capture user behavior and interactions required a thorough understanding of the business context and effective categorization. Class Imbalance: Addressing the class imbalance issue in predicting customer churn was crucial for model performance and reliability. Model Diversity: Ensuring diverse and independent base models for the ensemble approach to enhance predictive accuracy and robustness.


Advanced EDA: Exploratory Data Analysis provided insights into user registration trends and identified key factors influencing churn, guiding feature selection and model development. Feature Categorization: Effective feature engineering categorized data into Membership, User Log, and Transactional features, enabling a nuanced understanding of user behavior and interactions. Ensemble Modeling: Utilizing an ensemble approach with ten diverse base models from different feature sets mitigated class imbalance issues, enhancing model performance. Time-Dependent Analysis: Implementing time windows in feature generation ensured capturing variations in user behavior, making predictions more dynamic and reflective of temporal patterns.


Auto-renew and cancellation ratios emerged as key churn indicators, highlighting the pivotal role of transactional attributes. The final ensemble model excelled with a remarkable 96% accuracy and 86.5% F1-score, surpassing individual models and showcasing robustness. Transactional features, encompassing discounts, payment types, and subscription plans, outweighed user activity in predicting churn. In a saturated market, where quality is uniform among service providers, user activity became an unreliable predictor. The findings underscore the imperative for transaction-focused retention strategies in such competitive landscapes.