Predicting Customer Churn for Avast Mobile Using Historical Data

This thesis develops a propensity modeling framework to predict next-day churn among Avast Mobile Security users. Using behavioral, and notification data from the previous 7 days, and historical purchase data. The study trains two machine learning models, Logistic Regression and XGBoost, to generate churn propensity scores. Rather than making binary predictions, the models rank users by their likelihood to churn, enabling Avast to prioritize retention actions on the top 1% most at-risk users.