Survival analysis for predicting fitness app user churn

Fitness applications are increasingly used to support physical activity and promote healthier lifestyles. However, maintaining long-term engagement remains a major challenge, as many users discontinue app use within weeks. While churn prediction has been studied in fitness centers or other industries, research on digital fitness apps is still limited and often relies on static models such as logistic regression. To address this gap, this study analyses user churn in fitness apps using survival analysis techniques to identify factors contributing to drop out, aiming to improve user engagement and retention strategies. The study objective is to assess the suitability of survival analysis for predicting user churn times in fitness applications.

The study analyzed data from 3,034 users of the Mammoth Hunters fitness application. Three distinct time-range approaches were employed for survival analysis, each paired with two censoring methods. Kaplan-Meier estimates assessed user dropout probabilities over time, supplemented by parametric survival models and cure fraction models. Model performance was evaluated using mean absolute error, Akaike Information Criterion (AIC), concordance index, and Cox-Snell residuals.

Significant differences in retention were observed for multiple variables such as gender, activity level, training frequency, and body fat percentage (P=0.004) across all approaches. Men, older users, and those with higher training frequency showed longer engagement, while sedentary users and women disengaged earlier. LogNormal parametric models achieved the best predictive performance with mean absolute errors of 1.02, 1.94, and 3.32 weeks across time approaches. Cure models indicated that only a small fraction of users would remain engaged indefinitely.

This study highlights key factors driving user churn in the Mammoth Hunters fitness app, offering insights to help developers reduce dropout rates, enhance engagement, and improve user retention. Applying advanced survival and cure models can improve personalization, reduce dropout rates, and support sustainable health outcomes through digital fitness platforms.