Survival analysis for predicting fitness app user churn

Area of Science:

• Digital Health
• Health Informatics
• Behavioral Science

Background:

• Fitness applications are widely adopted for promoting physical activity and healthy lifestyles.
• Long-term user engagement is a significant challenge, with high dropout rates within weeks of adoption.
• Existing churn prediction research in digital fitness apps is limited, often using basic statistical models.

Purpose of the Study:

• To analyze user churn in fitness applications using survival analysis techniques.
• To identify key factors contributing to user dropout in digital fitness apps.
• To assess the suitability of survival analysis for predicting user churn times and improving retention strategies.

Main Methods:

• Survival analysis was applied to data from 3,034 users of the Mammoth Hunters fitness application.
• Multiple survival analysis approaches were used, including Kaplan-Meier, parametric models, and cure fraction models.
• Model performance was evaluated using metrics like mean absolute error and concordance index.

Main Results:

• Significant differences in retention were observed based on gender, age, activity level, and training frequency.
• Men, older users, and those with higher training frequency demonstrated longer engagement.
• LogNormal parametric models provided the best predictive performance for user churn.

Conclusions:

• Survival and cure models offer valuable insights into user churn dynamics in fitness apps.
• Identifying key factors can help developers enhance personalization, reduce dropout rates, and improve user retention.
• Advanced modeling can support sustainable health outcomes through improved digital fitness platforms.