Beyond Binary Classification: Time-to-Event Modeling for Player Retention Using Cox Proportional Hazards and Ensemble Learning

Md. Wira Putra Dananjaya

doi:10.56427/jcbd.v5i1.792

Authors

Md. Wira Putra Dananjaya Universitas Pendidikan Nasional

DOI:

https://doi.org/10.56427/jcbd.v5i1.792

Keywords:

Churn Prediction, Survival Analysis, Cox Proportional Hazards, XGBoost, Game Analytics

Abstract

Player retention is the primary economic driver in the Free-to-Play (F2P) gaming industry, yet traditional churn prediction methodologies often rely solely on binary classification, neglecting the critical temporal dimension of when a player is likely to leave. Furthermore, the scarcity of open-source behavioral datasets restricts the development of reproducible academic frameworks. This study addresses these gaps by proposing a hybrid analytical framework that integrates Ensemble Learning (XGBoost) for predictive precision and Survival Analysis (Cox Proportional Hazards) for time-to-event risk modeling, utilizing a realistically simulated dataset of 5,000 players. Experimental results indicate that while the XGBoost model achieves robust discriminative stability with an AUC of roughly 0.90, the Survival Analysis provides deeper explanatory insights, revealing that game progression (level_reached) is a significantly more dominant determinant of retention (Hazard Ratio < 1.0) than short-term recency metrics. These findings suggest that depth of commitment acts as a stronger buffer against churn than login frequency, offering game developers a quantifiable basis to shift retention strategies from generic daily incentives to progression-based milestones. By openly providing the simulated dataset and full analytical pipeline, this work also contributes a reproducible methodological template for future game analytics research in data-scarce environments.

Downloads

Download data is not yet available.

References

K. Mustač, “Predicting Player Churn of a Free-to-Play Mobile Video Game Using Supervised Machine Learning,” Appl. Sci. Switz., vol. 12, no. 6, 2022, doi: 10.3390/app12062795.

J. B. Débordès, “Is my cross-promotion profitable? Evaluation of game-to-game cannibalization in free-to-play mobile games,” J. Mark. Anal., vol. 9, no. 3, pp. 173–184, 2021, doi: 10.1057/s41270-021-00122-x.

S. Abdul-Rahman, “Enhancing churn forecasting with sentiment analysis of steam reviews,” Soc. Netw. Anal. Min., vol. 14, no. 1, 2024, doi: 10.1007/s13278-024-01337-3.

E. Akhmetshin, “Intelligent Data Analytics using Hybrid Gradient Optimization Algorithm with Machine Learning Model for Customer Churn Prediction,” Fusion Pract. Appl., vol. 14, no. 2, pp. 159–171, 2024, doi: 10.54216/FPA.140213.

C. Lukita, “Predictive and Analytics using Data Mining and Machine Learning for Customer Churn Prediction,” J. Appl. Data Sci., vol. 4, no. 4, pp. 454–465, 2023, doi: 10.47738/jads.v4i4.131.

M. P. Bach, “Churn Management in Telecommunications: Hybrid Approach Using Cluster Analysis and Decision Trees,” J. Risk Financ. Manag., vol. 14, no. 11, 2021, doi: 10.3390/jrfm14110544.

N. Almufadi, “Deep Convolutional Neural Network Based Churn Prediction for Telecommunication Industry,” Comput. Syst. Sci. Eng., vol. 43, no. 3, pp. 1255–1270, 2022, doi: 10.32604/csse.2022.025029.

K. W. D. Bock, “Spline-rule ensemble classifiers with structured sparsity regularization for interpretable customer churn modeling,” Decis. Support Syst., vol. 150, no. Query date: 2025-12-02 00:22:28, 2021, doi: 10.1016/j.dss.2021.113523.

H. Kang, “Match experiences affect interest: Impacts of matchmaking and performance on churn in a competitive game,” Heliyon, vol. 10, no. 3, 2024, doi: 10.1016/j.heliyon.2024.e24891.

M. Bogaert, “Ensemble Methods in Customer Churn Prediction: A Comparative Analysis of the State-of-the-Art,” Mathematics, vol. 11, no. 5, 2023, doi: 10.3390/math11051137.

A. Appenzeller, “Privacy and Utility of Private Synthetic Data for Medical Data Analyses,” Appl. Sci. Switz., vol. 12, no. 23, 2022, doi: 10.3390/app122312320.

R. Flunger, “Game Analytics on Free to Play,” Commun. Comput. Inf. Sci., vol. 1054, no. Query date: 2025-06-04 16:23:08, pp. 133–141, 2019, doi: 10.1007/978-3-030-27355-2_10.

D. C. Feng, “Interpretable XGBoost-SHAP Machine-Learning Model for Shear Strength Prediction of Squat RC Walls,” J. Struct. Eng. U. S., vol. 147, no. 11, 2021, doi: 10.1061/(ASCE)ST.1943-541X.0003115.

Z. Alfasanah, M. Niam, S. Wardiani, M. Ahsan, and M. Lee, “Monitoring air quality index with EWMA and individual charts using XGBoost and SVR residuals,” MethodsX, no. Query date: 2025-11-25 10:19:06, 2025, [Online]. Available: https://www.sciencedirect.com/science/article/pii/S2215016124005582

V. Baralou, “Individual risk prediction: Comparing random forests with Cox proportional-hazards model by a simulation study,” Biom. J., vol. 65, no. 6, 2023, doi: 10.1002/bimj.202100380.

A. A. Ewees, “Boosting arithmetic optimization algorithm with genetic algorithm operators for feature selection: Case study on cox proportional hazards model,” Mathematics, vol. 9, no. 18, 2021, doi: 10.3390/math9182321.

A. Drachen, “Going out of business: Auction house behavior in the Massively Multi-player Online Game Glitch,” Entertain. Comput., vol. 5, no. 4, pp. 219–232, 2014, doi: 10.1016/j.entcom.2014.09.001.