ML Classification: Predicting Audiobook App User Retention

This project applies supervised machine learning techniques to predict user churn in a commercial audiobook application. Using a dataset of over 14,000 anonymized user records, we developed a classification pipeline that identifies at-risk users based on behavioral and transactional features. Surprisingly, users with higher engagement metrics—such as audiobook completion and listening time—were more likely to churn, revealing a “one-and-done” behavioral pattern. After addressing class imbalance with SMOTE and evaluating multiple models, a tuned Random Forest classifier achieved 88% precision and 82% F1-score on the active class. The findings challenge conventional engagement metrics and offer actionable insights for retention strategies based on post-purchase behavior rather than content consumption.