Predicting User Dropout in Digital Health

Goal: Use machine learning to help a digital health company undestand what users are most likely to drop out, what factors might be associated with dropout, and what time points might be important for intervention.

Methods: We used regression analysis, cross-validated gradient boosted decision tree and survival analysis to explore the dataset.

Findings: User age, month of enrollment, frequency of activity in the first two weeks are strong predictors of user dropout within the first 180 days. Survival analysis confirmed that 180 days was an important time point. 

Most Important Features Associated with User Dropout
Most Important Features Associated with User Dropout

Last updated:

December 2, 2017