Early Churn Prediction from Large Scale User-Product Interaction Time Series
Published in International Conference on Machine Learning and Applications (ICMLA), IEEE, 2023
A deep-learning model that ingests raw user-product interaction time series (rather than aggregated session features) to surface churn signal weeks before conventional baselines. Demonstrated on a 250M+ user feed in a production fantasy-sports setting. The architecture captures temporal cadence and ordering of events that flattened features discard, and trades off lead time vs. precision via a tunable horizon head.
Recommended citation: Patil, N. et al. (2023). Early Churn Prediction from Large Scale User-Product Interaction Time Series. 2023 International Conference on Machine Learning and Applications (ICMLA). IEEE.
Download Paper