Predictive Modeling of Online Game User Retention Based on Flow Theory and LightGBM

Abstract

With the global game industry entering a new stage of stock competition and experience value driven, accurate prediction of user retention trend has become the core proposition of refined operation of enterprises. In order to solve the problem that the behavior data is disconnected from the players' psychological state in online game retention prediction, this paper proposes a prediction framework that combines psychological flow theory and LightGBM algorithm. By constructing the quantitative index FlowIndicator of cardiac flow imbalance threshold, the qualitative psychological experience is transformed into a dynamic tracking numerical system. Based on the behavior logs of 40000 MMORPG players, using LightGBM algorithm and SHAP interpretability tool, a retention prediction system with time series adaptability is constructed. The experimental results show that the AUC of the model test set reaches 0.93 and the F1 score reaches 0.89, which significantly improves the recognition accuracy of the recessive loss signal and provides interdisciplinary methodological support for the refined operation of the game.