Аннотация
Modern gamified EdTech platforms require a systematic scientific approach to analyzing large-scale user data. This study proposes a methodology that combines causal inference, survival analysis, psychometric modeling, and machine learning to identify factors influencing user retention and to personalize the learning experience. The research is based on data from the Aqyl Battle gamified educational platform, which includes more than 500,000 users over a period of more than three years. Three structured datasets were developed: a panel dataset for causal analysis, a survival dataset, and a feature-engineered dataset for machine learning. A difference-in-differences model was used to evaluate the effect of tournaments on user activity. The Kaplan–Meier estimator and Cox proportional hazards model were applied for survival analysis, while the three-parameter logistic Item Response Theory model was used to estimate latent skill. Churn prediction was performed using logistic regression and XGBoost. The results show that the introduction of tournaments increased weekly user activity by 12.4% (p < 0.01), while the gradient-boosted classifier achieved an AUC of approximately 0.87 on a temporally separated test set. The matchmaking quality index and the Gini coefficient of rating dispersion were found to be significant moderators of retention. When match balance fell below 0.6, weekly attrition increased sharply. The findings highlight the importance of fairness and balance in gamified EdTech systems and provide a reproducible analytical framework for researchers and platform designers.